Projects
| # | Title | Team Members | TA | Professor | Documents | Sponsor |
|---|---|---|---|---|---|---|
| 1 | AI Facial Recognition for Automated Room Access |
Chaohua Yao Haowen Lin Jianchong Chen Zitong Qu |
design_document1.pdf other1.pdf other2.docx |
Hua Chen | ||
| PROBLEM Traditional check-in is slow, labor-intensive, and insecure. Manual verification creates bottlenecks, and physical cards are easily lost or shared. SOLUTION OVERVIEW An AI-powered kiosk that automates room card issuance. If the users registered ; the kiosk then uses facial recognition detection to verify identity and instantly dispense an stundent room card. SUBSYSTEMS Subsystem 1: Kiosk Hardware & Interface: Designing the terminal and mechanical dispenser. Subsystem 2: Facial Biometrics: Implementing CNN-based recognition and anti-spoofing liveness detection. Subsystem 3: Mobile Registration: A secure palce for user photo and ID uploads to a cloud database. Subsystem 4: Authorization Logic: The algorithm that matches live scans to the database to trigger card issuance. CRITERION FOR SUCCESS Verification and issuance completed Zero manual staff intervention required for 24/7 operation. High-accuracy biometric binding to prevent unauthorized access. DISTRIBUTION OF WORK Jianchong Chen :Facial biometrics and liveness detection Zitong Qu:Registration and database management. Chaohua Yao: Kiosk hardware integration. Haowen Lin:Authorization algorithm and security protocols. |
||||||
| 2 | Movable Impact Testing Platform |
Bingkun Fu Feiyu Tang Shangyu Wang Yihang Shen |
design_document1.pdf final_paper1.pdf final_paper2.pdf video |
Binbin Li | ||
| This project develops a movable impact testing platform capable of generating controllable impact forces and tunable impact frequencies. The system adjusts impact height, contact tips and actuation frequency through a mechanical transmission mechanism, producing repeatable and measurable impact forces with varying intensities. The platform is intended for rapid dynamic testing of bridge structures and similar civil infrastructures, supporting studies in structural dynamics and structural health monitoring. | ||||||
| 3 | Wearable mobility-assistance device for Blind and visually impaired (BVI) |
Darui Xu Haoyu Zhu Jiashen Ren Jinnan Zhang |
design_document1.pdf final_paper1.pdf final_paper2.pdf |
Bo Zhao | ||
| # Problem Blind and visually impaired (BVI) individuals rely heavily on hearing to navigate safely in daily environments. While walking, they must continuously monitor critical environmental sounds such as approaching vehicles, crosswalk signals, bicycles, and nearby pedestrians. However, many existing assistive navigation devices communicate obstacle information mainly through audio alerts or voice prompts. This creates a major usability and safety issue because the device competes with the same auditory channel that the user depends on for situational awareness. In addition, audio-based systems often require earphones or louder playback in noisy environments, which can further reduce a user’s ability to perceive surrounding hazards. As a result, these systems may unintentionally compromise safety instead of improving it. The problem addressed by this project is therefore: **how to provide intuitive and timely obstacle-location information to BVI users without occupying their auditory channel**. This problem is important because an effective mobility-assistance device must do more than detect obstacles. It must communicate actionable information in a way that is fast, intuitive, wearable, and compatible with the user’s natural navigation behavior. Our project focuses on preserving hearing for environmental awareness while shifting obstacle communication to the tactile channel. # Solution Overview This project proposes a **wearable haptic navigation-assistance device** for blind and visually impaired users. The system will detect nearby obstacles in front of the user using an AI vision-based sensing approach and communicate their relative direction and distance through **vibration feedback** rather than sound or speech. The proposed system will use a camera and onboard processing hardware to capture visual information from the environment. AI-based vision algorithms will analyze the scene in real time to identify nearby obstacles and estimate their relative position with respect to the user. Based on this information, the system will activate vibration motors to convey obstacle direction and distance. For example, vibration on the left side may indicate an obstacle on the left, while stronger or faster vibration may indicate a closer obstacle. The key innovation of this design is a **non-auditory feedback mapping** that allows users to receive obstacle information while keeping their hearing fully available for environmental sounds. Compared with conventional audio-based systems, this approach is intended to improve safety, reduce sensory conflict, and provide a more intuitive navigation aid in realistic walking scenarios. To keep the project feasible within the course scope, the prototype will focus on short-range obstacle awareness and vibration-based haptic communication rather than full autonomous navigation or large-scale scene understanding. # Components The system will be organized into the following major subsystems: ## AI Vision Sensing Subsystem The sensing subsystem detects nearby obstacles and estimates their relative position using visual data. Possible components include: - Camera module - Embedded AI processing unit or microprocessor - Computer vision / object detection algorithm - Distance or relative-position estimation logic This subsystem is responsible for acquiring environmental information and identifying obstacles in real time. ## Processing and Control Subsystem The processing subsystem interprets the sensing results and determines the correct haptic response. Possible components include: - Embedded controller or processor - Obstacle localization logic - Haptic feedback mapping algorithm - Timing and control logic This subsystem converts vision-based obstacle information into control commands for the feedback device. ## Vibration Feedback Subsystem The feedback subsystem communicates obstacle information to the user through tactile vibration cues. Possible components include: - Vibration motors - Motor driver circuitry - Wearable actuator placement - Feedback mapping design for direction and distance This subsystem is responsible for conveying obstacle direction and relative distance in an intuitive and distinguishable way. ## Power Subsystem The power subsystem provides portable and stable power to all electronics. Possible components include: - Rechargeable battery - Voltage regulation circuit - Charging interface - Power switch and protection circuitry This subsystem enables continuous wearable operation. ## Wearable Integration Subsystem The wearable integration subsystem packages the prototype into a form suitable for real use. Possible components include: - Wearable mounting structure - Sensor and actuator supports - Wiring and enclosure management - Adjustable fastening mechanism This subsystem ensures that the device is practical, lightweight, and wearable. # Criteria of Success The project will be considered successful if the final prototype satisfies the following criteria: 1. The device must detect nearby obstacles within the intended range with reliable performance during indoor testing. 2. The system must communicate obstacle direction and relative distance through vibration feedback in a way that users can correctly interpret during controlled testing. 3. The device must provide obstacle information without using audio output, thereby preserving the user’s auditory awareness of the surrounding environment. 4. The final prototype must function as a wearable, battery-powered system capable of real-time operation during demonstration. |
||||||
| 4 | SoftReach Arm |
Jinwen Wang Junyi Chen Ruxi Deng Zhian Xie |
design_document1.pdf final_paper1.pdf other1.pdf proposal1.pdf |
Shi Ye | ||
| # SoftReach Arm SoftReach Arm is a robotic system that combines **vision-based object detection** with a **soft manipulator** for object grasping. ## Project Goal The goal of this project is to evaluate the benefits of **compliance** in robotic manipulation, particularly when dealing with uncertainty in: - object pose - object geometry By integrating visual perception with a compliant soft arm, the system aims to improve robustness and adaptability during grasping tasks. ## Actuation Method The soft manipulator will be driven by **pneumatic actuation**. Pneumatic systems can generate fast and large deformations while maintaining a relatively simple structure, making them suitable for **soft robotic manipulation**. In addition, pneumatic actuation is convenient to implement within a **short-term capstone project**, allowing the system to be built and tested efficiently. |
||||||
| 5 | An event-based smart vision node for ultra-low-latency motion detection |
Luying Wang Shuke Wang Yaxing Zhang Yueyao Si |
design_document1.pdf final_paper1.pdf final_paper2.pdf proposal1.docx |
Aili Wang | ||
| # Problem Traditional motion detection systems usually rely on frame-based cameras, which capture full images at fixed intervals. In many situations, consecutive frames are very similar, making the system store and process a large amount of redundant information. This not only increases data load but also leads to higher power consumption. Meanwhile, in this way, motion can only be analyzed after a batch of image frames are collected and processed, which is not ideal for applications that require very low latency. As a result, the main problem is how to build a vision system that can respond to motion more efficiently by using only meaningful visual changes instead of full frames, while showing potential advantages in latency, resources and power consumption compared with a conventional approach. # Solution Overview Our solution is to build an event-based vision system using a DVS camera, FPGA, and SNN-inspired processing. Instead of capturing and processing full image frames, the system works directly on event data input stream. The system first captures event-based visual data from a DVS camera. These events are then sent to the FPGA, where they are received, parsed, and temporarily buffered in real time without reconstructing full frames. The formatted event stream is then passed to a software-based SNN-inspired module, which analyzes motion patterns over time and generates a detection result when meaningful activity is observed. When motion is detected, the result will be sent to the output subsystem for display with minimal latency. If time allows, a frame-focused baseline may be used as a comparison so that our system can be evaluated in terms of end-to-end latency, event throughput, and power consumption. # Solution Components & Distribution of work ### Event-Based Vision Sensor (Shuke Wang – EE) - Dynamic Vision Sensor (DVS) Camera: Employs a neuromorphic event-based sensor that captures visual information asynchronous spikes of pixel-level brightness changes. Each event includes pixel coordinates, polarity, and a precise microsecond timestamp, enabling ultra‑low‑latency motion detection without the need for full frame readout. - High‑Speed Data Interface: Outputs event streams using the Address‑Event Representation (AER) protocol over a high‑bandwidth link. This interface allows direct, real‑time transmission of raw events to the FPGA processing platform, minimizing additional latency, and preserving the temporal precision of the sensor. - Optics and Mounting: The camera is equipped with a suitable lens to match the target field of view and application scenario. It is rigidly mounted on an adjustable stage to facilitate precise alignment and stable imaging conditions during experiments. ### FPGA Subsystem (Yaxing Zhang – EE) - The FPGA subsystem serves as the real-time processing platform of the system. It receives the event stream from the DVS camera through a high-speed interface and parses each event into pixel coordinates, polarity, and timestamp. - The parsed events are temporarily stored in on-chip buffers to maintain stable data flow and handle burst event traffic. The FPGA can also perform lightweight pre-processing such as basic filtering before passing the formatted event stream to the motion detection module. - This hardware platform ensures low latency and efficient handling of asynchronous event data in the system pipeline. ## SNN-Based Motion Detection Subsystem (Luying Wang – ECE) - An SNN-inspired module that analyzes incoming events, detects motion regions by updating neural activity based on event spikes, builds up motion activity in certain regions, and generates an output when the activity exceeds a threshold. ### Output Subsystem (Yueyao Si – ME) - The output subsystem is responsible for presenting the final motion detection result generated by the SNN-inspired module. Once motion activity exceeds the predefined threshold, a detection signal is produced and forwarded to the output controller. - In the current implementation, the FPGA receives the detection result and triggers a visual indicator such as an LED or display module. When motion is detected, the indicator is activated in real time; otherwise it remains off. - This subsystem provides a simple and low-latency way to demonstrate the system response to motion events. The output interface can also be extended to support other devices, such as a monitor display, UART logging interface, or external control signals for robotic or embedded applications. # Criteria of Success ### Functionality - The complete pipeline runs successfully from event input to final output. - The motion detection module can correctly identify motion regions from the event stream. - The output responds correctly to motion: the display turns on when motion is detected and remains off otherwise. ### Performance - The end-to-end latency is less than 50 ms. - The measured FPGA board power during operation is less than 5 W. - The FPGA resource utilization remains below 80% of available logic and memory resources. # References - [Event-based Vision: A Survey](https://arxiv.org/pdf/1904.08405) - [Event-based vision on FPGAs – a survey The work presented in this paper was supported by: the program ”Excellence initiative –- research university” for the AGH University of Krakow.](https://arxiv.org/html/2407.08356v1#bib.bib60) - [Neuro-Inspired Spike-Based Motion: From Dynamic Vision Sensor to Robot Motor Open-Loop Control through Spike-VITE](https://www.mdpi.com/1424-8220/13/11/15805) - [A Reconfigurable Architecture for Real-time Event-based Multi-Object Tracking | ACM Transactions on Reconfigurable Technology and Systems](https://dl.acm.org/doi/10.1145/3593587) |
||||||
| 6 | Automated Intelligent Document Stamping System with Machine Vision Integration |
Jiaheng Zeng Peter Chen Xuliang Huang Zhiqiang Qiu |
appendix1.pdf design_document1.docx design_document2.pdf final_paper1.pdf other1.pdf photo1.jpg proposal1.pdf video |
Fangwei Shao | ||
| This project aims to design and build a mechatronic system capable of automating the manual process of stamping multi-page documents. By integrating UI interface, machine vision, and an automatic page-turner, the system will eliminate repetitive labor and ensure precise and convenient stamp placement. The main work of mechanical task is to build X-Y positioning arms so the stamp head can move on the XY plane and press at the target location. And paper feeding and output path are also required. Similar to a printer, a pickup roller brings one sheet to the stage for stamp. After stamping, an output roller pushes the sheet to the exit tray. In the UI interface, the users can learn how to use this machine. And then, the users may choose different document stamping modes in the UI interface (we may develop different modes based on the demands). Functional Requirements: 1. Motion Control: • XY-axis movement to cover standard or non-standard page sizes (usually A4/Letter). • Z-axis pressure detection for the adjustable stamping motion. 2. Vision & Logic: • Mode A (Content-Aware): Detect text content and stamp in appropriate open spaces. • Mode B (Keyword): Recognize specific characters (OCR) and stamp at a relative offset. • Combine a VLM and a camera to achieve stamping location detection. 3. Pages Handling: • Separate paper to ensure feeding only one paper into the stamping stage and withdrawing paper from the stamping stage. The whole process should avoid jamming and be transparent to the users for validation. 4. User Interaction: • Software interface to select modes, input keywords, and monitor progress. Goal: Ultimately, we hope this project can help our ZJUI administrative staffs to liberate their hands from repetitive stamping tasks, saving their time. |
||||||
| 7 | Motion Analysis and Trajectory Reconstruction of Smart SoftBall with UWB Positioning and Inertial Sensing |
Chenhan Yang Tianyang Sun Yuxing Wu |
design_document1.docx final_paper1.pdf final_paper2.pdf photo1.png photo2.jpg presentation1.pptx |
|||
| # Problem Traditional softball and tennis games or training rely on human umpires or cameras to determine the ball's landing point and status. Human judgment is susceptible to error, resulting in low efficiency in training and game statistics. Camera-assisted systems (such as Hawk-Eye) require 3D modeling of the venue, at least 8-10 ultra-high-speed cameras, and a range of other equipment. This is costly, requires fixed locations, and is not mobile, making it difficult to widely apply to daily training or amateur matches. Furthermore, it cannot record the ball's speed, and trajectory in real time, resulting in incomplete training data. # Solution overview This project proposes a general-purpose intelligent softball system based on UWB: 1. A miniature UWB transmitter is embedded in the ball to transmit positioning signals in real time. 2. UWB receivers are placed at the four corners of the field to calculate the ball's three-dimensional coordinates using the TDoA algorithm. 3. Using fusion algorithms such as Kalman filtering combined with UWB data, the system can restore the ball's continuous trajectory and velocity direction. 4. The system can determine in real time whether the landing point is out of bounds, providing training and game data analysis. # Solution component 1. **Ball module: Miniature UWB transmitter + ultra-small battery ** 2. **Lightweight Design for Guaranteed Ball Flight Performance** 3. **Field Receiver Module:** * Four or more UWB receivers, fixed at the four corners of the court. * Receives ball-transmitted signals and calculates TDoA (Total DoA). 4. **Data Processing and Fusion Algorithm:** * Uses Kalman filtering and UWB data fusion. * Reconstructs the ball's 3D trajectory, axis of rotation, and rotational speed. 5. **Visualization and Analysis Platform:** * Real-time trajectory display. * Landing point and out-of-bounds determination. * Training data statistics (number of hits, speed, spin). * Report generation and technical improvement suggestions. # Criterion for Success The evaluation criteria for project success include: 1. **Hardware Implementation:** The embedded UWB module in the ball functions normally without affecting the ball's flight performance. 2. **Positioning Accuracy:** Using the four corner receivers, centimeter-level 3D position determination is achieved. 3. **Trajectory and Spin Recovery:** Continuous trajectory and rotational angular velocity can be accurately reconstructed. 4. **Real-time Out-of-Bounds Determination:** The system can automatically determine whether the ball is out of bounds, and the error is ≤ 5 cm compared with the accuracy of manual determination. 5. Data Statistics and Visualization: The system can generate training data such as ball speed, spin, and landing point, and display it visually. 6. Scalability: The system is suitable for multi-venue, mobile deployment, and is ideal for training and recreational matches. |
||||||
| 8 | Guided Robotic Manipulator for Chinese Calligraphy |
Nuoer Huang Xinyi Shen Xirui Yao Yujie Wei |
design_document1.pdf final_paper1.pdf final_paper2.docx other1.docx presentation1.pptx video1.mp4 |
Meng Zhang | ||
| #### PROBLEM: Traditional Chinese calligraphy is a sophisticated art form that demands meticulous control over three-dimensional movement, tip pressure (stroke depth), and fluid velocity. Currently, most robotic manipulators are designed for rigid industrial tasks and struggle to replicate the "softness" of a brush and the nuanced transitions required for different artistic styles like Kaishu (Regular Script) or Cursive. There is a need for a high-precision system that can translate the aesthetic essence of human calligraphy into robotic motion, preserving cultural heritage through modern ECE technologies. #### SOLUTION OVERVIEW: The project aims to design an intelligent robotic manipulator system capable of executing complex Chinese calligraphy. The system provides real-time control over the brush's trajectory and pressing force to achieve varying stroke widths and styles. By integrating advanced motor control algorithms and sensor feedback, the system coordinates multi-axis movements to achieve the necessary stroke dynamics and tip positioning required for traditional calligraphy. It processes digital stroke data and translates it into precise mechanical movements, ensuring the robotic arm can interact with the paper surface with artistic fidelity. #### SOLUTION COMPONENTS: ####Manipulator Hardware Modules: -A multi-DOF mechanical structure designed to provide the necessary range of motion and stability for complex characters. -Utilizing high-torque servo motors to ensure smooth and accurate movement of each joint. -A specially designed brush holder with an integrated damping or spring mechanism to simulate the flexibility of a human hand. ####Control & Processing Modules: -A software module used to convert vector-based character paths (SVG/G-code) into synchronized motor angles using inverse kinematics. -A microcontroller or PC-based system that processes stroke data and coordinates real-time motion commands. -A feedback loop that manages the Z-axis height to control the contact area between the brush tip and the paper. ####Sensing & Perception Modules: -Utilizing sensors (such as FSR) to monitor the pressure exerted on the paper in real-time to prevent paper damage. -Using a camera to calibrate the initial position of the brush and paper, ensuring the writing stays within the designated boundaries. #### CRITERION FOR SUCCESS: -Artistic Fidelity: The system must replicate standard characters with a spatial deviation of less than 2mm compared to the digital template. -Stroke Variation: The manipulator must demonstrate the ability to produce varying stroke thicknesses by dynamically adjusting the Z-axis depth. -Operational Stability: The system should be able to perform multiple consecutive writing tasks without mechanical failure or recalibration. -System Integration: The hardware and software must maintain a stable connection with low latency during real-time command transmission. #### DISTRIBUTION OF WORK: - ECE STUDENT WEI YUJIE: Leading the design and assembly of the robotic arm structure and the custom brush end-effector. Responsible for the physical stability of the manipulator and the selection of mechanical components. -ECE STUDENT YAO XIRUI: Implementing the motor driver circuits and the low-level firmware for real-time motor control. Ensuring precise execution of the joint angles provided by the server. -EE STUDENT SHEN XINYI: Developing the inverse kinematics (IK) model and trajectory planning algorithms. Converting character stroke data into smooth, coordinated motion paths for the arm. -ME STUDENT HUANG NUOER: Implementing the vision-based calibration system and the user interface. Responsible for system-wide integration, data transmission between the server and hardware, and pressure sensor feedback logic. |
||||||
| 9 | Spherical Bio-Inspired Tensegrity Multiple Step Robot with LCE Actuation |
Dongzi Li Yiqin Xiang Yuxuan Huang Ziye Chen |
final_paper1.pdf other1.pdf |
Hanzhi Ma | ||
| Problem Robots designed for exploration or operation in unstructured environments often face challenges related to impact resistance, adaptability, and mechanical complexity. Traditional rigid robots rely on wheels, motors, and rigid frames, which can be vulnerable to collisions and mechanical damage when operating in uncertain environments. Tensegrity structures offer a promising alternative because they combine rigid compression elements and tensioned cables to form lightweight yet resilient structures. These systems distribute loads efficiently and naturally absorb impacts, making them suitable for robots that must tolerate collisions or uneven terrain. However, many existing tensegrity robots rely on bulky motors or external actuation systems to control cable tension, which increases system weight, mechanical complexity, and power consumption. A more compact and integrated actuation method is needed to enable lightweight tensegrity robots capable of controlled locomotion. Researchers therefore need a robotic system that integrates lightweight tensegrity structures with compact actuators to enable controllable motion while maintaining structural compliance and robustness. Solution Overview Our solution is to develop a spherical bio-inspired tensegrity robot that uses liquid crystal elastomer (LCE) actuator cables as artificial muscles. The robot will consist of a lightweight hollow tensegrity framework composed of rigid rods connected by tension cables. Selected cables will be replaced by LCE actuators. When electrical current is applied to the LCE cables, Joule heating causes the material to contract. This contraction increases tension within the tensegrity structure and produces controlled deformation of the spherical frame. By activating different actuators in sequence, the robot can shift its center of mass and generate rolling motion on flat ground. The system will include a multi-channel control circuit, wireless communication, and gait-sequence control softwarethat coordinate actuator activation. These components allow the robot to perform continuous locomotion and basic directional control. In addition, power and thermal management mechanisms will ensure safe and reliable operation of the LCE actuators. Solution Components Actuation Subsystem • LCE actuator cables that contract when electrically heated • Driver circuits capable of supplying current to multiple actuators • Electrical connections integrated into the tensegrity structure These components act as artificial muscles that control tension within the robot and enable structural deformation. Structural Subsystem • Lightweight rigid rods forming the compression elements • Tension cables connecting structural nodes • Mechanical joints and connectors forming a spherical tensegrity framework The structure maintains the robot’s geometry while distributing loads and absorbing impacts. Control Subsystem • Microcontroller for actuator control • Multi-channel switching or driver circuitry • Gait-sequence control software This subsystem determines the timing and sequence of actuator activation required to generate rolling locomotion. Communication Subsystem • Wireless communication module (e.g., Bluetooth or WiFi) • Remote command interface This subsystem allows users to send commands to control the robot’s motion. Power Subsystem • Rechargeable battery pack • Voltage regulation circuitry This subsystem supplies electrical power to the control electronics and LCE actuators. Criterion for Success The project will be considered successful if the following criteria are achieved: 1. Actuation capability The LCE actuator cables must demonstrate repeatable contraction when electrically heated and return to their original length after cooling. 2. Controlled locomotion The robot must demonstrate continuous rolling motion on a flat surface through coordinated actuator activation. 3. Directional control The robot must demonstrate basic steering capability using different actuator sequences. 4. Wireless operation The robot must successfully receive wireless commands and execute corresponding motion behaviors. 5. System integration The complete system—including tensegrity structure, actuators, control electronics, and power supply—must operate together as an integrated robot. |
||||||
| 10 | Latte Art Coffee Machine |
Jingyang Cao Tianheng Wu Yuhao Shen Yukang Lin |
design_document1.pdf final_paper1.pdf final_paper2.pdf presentation1.pptx proposal2.docx video1.mp4 |
Wee-Liat Ong | ||
| Our project aims to develop a **Latte Art Coffee Machine** that can automatically print simple latte art patterns on the surface of coffee based on user-uploaded images. ## Problem Traditional latte art requires significant manual skill and practice, which makes it difficult for ordinary users or small coffee shops to consistently create customized patterns. Existing coffee machines usually focus on brewing quality rather than personalized visual presentation. We want to solve the problem of making latte art more accessible, repeatable, and customizable. ## Solution Overview Our proposed solution is a coffee machine add-on/system that accepts a user-uploaded image, extracts its simple outline through basic image processing, and converts the outline into motion paths. These motion paths will then drive an **X-Y motion platform** so that the machine can print patterns onto the coffee surface with acceptable accuracy and repeatability. ## Components The system will be divided into several main components: - **Image Processing Module** Processes user-uploaded images and extracts simplified outlines suitable for printing. - **Path Planning / Control Module** Converts image outlines into printable motion trajectories and generates control signals. - **X-Y Motion Platform** Uses stepper motors and a mechanical positioning system to move the printing head accurately over the coffee surface. - **Dispensing / Printing Mechanism** Deposits the latte art material onto the foam surface according to the planned path. - **Power and System Integration Module** Provides stable power and coordinates communication among all hardware and software modules. ## Criteria of Success A successful project should meet the following goals: 1. The system can accept a user-input image and generate a simplified printable outline. 2. The machine can convert the outline into valid motion commands for the X-Y platform. 3. The X-Y platform can move reliably and repeatedly to the desired positions using stepper motors. 4. The final system can print recognizable latte art patterns on the coffee surface. 5. The overall design is feasible, reliable, and safe for demonstration in an engineering project setting. ## Feasibility and Value This project combines image processing, motion control, embedded systems, and mechanical design in a practical and interesting application. It is challenging enough for a senior design project while still remaining feasible with modular implementation and step-by-step testing. |
||||||
| 11 | JengaBot - A Robotic System for Playing Jenga with Human |
Hengtie Zhu Jiacheng Ye Peiran Wei Wangyihan Guo |
design_document1.pdf design_document2.pdf final_paper1.pdf final_paper2.pdf |
Pavel Loskot | ||
| # JengaBot - A Robotic System for Playing Jenga with Human Team members (listed A-Z): - Wangyihan Guo - Peiran Wei - Jiacheng Ye - Hengtie Zhu ## Problem While technologies such as 3D printing and embodied intelligence have progressed significantly, precise object manipulation in constrained spaces continues to be a complex problem. Limited space restrict massive robotic approach yet allow 3D-priting frame to participate in the workflow, thus we dropped out this problem to discover the possibility of combing 3D-priting frame with robotic approaches. ## Solution Overview While confined spaces often restrict the deployment of bulky robotic systems, they present a unique opportunity to utilize customized 3D-printed frameworks. To address this spatial limitation, we dedicated our project to exploring the integration of compact 3D-printed structures with precision robotic mechanisms. To achieve this integration, we engineered a three-axis motion control system designed to maneuver an operation unit precisely within the restricted workspace. We demonstrated the capabilities of this system by programming the machine to play Jenga interactively against a human opponent, a task requiring delicate block removal without collapsing the tower. The success of this interactive system relies on continuous environmental monitoring. The real-time status of the Jenga tower is captured by four cameras strategically mounted at the corners of the 3D-printed framework, ensuring comprehensive visual coverage of the workspace from multiple angles. The visual data captured by these cameras is continuously fed into a Raspberry Pi for processing. Utilizing predefined algorithmic strategies, the Raspberry Pi analyzes the physical state of the tower and directly drives the operation unit to execute the optimal next move, completing the automated feedback loop. ## Solution Components - Three-axis motion control system: used to move the operation unit precisely in the constrained space. - Operation unit: a specialized operational unit capable of executing complex physical interactions with the Jenga tower, including grasping, pushing, and translating the bricks. - Camera: used to collect live visual data of the Jenga tower to Raspberry Pi. - Raspberry Pi: a central controller that constructs a real-time digital model of the Jenga tower from visual data, directing the end-effector to execute predefined algorithmic strategies. ## Criterion for Success 1. The system could play the game with human opponents with reasonable react. 2. The system could handle scenarios out of rules, e.g. human manually make tower collapse or try to make a second move in one round. ## Distribution of the Work Wangyihan Guo & Peiran Wei: Responsible for the controlling unit and cameras, including strategy written, the interface between software and hardware, etc. Jiacheng Ye & Hengtie Zhu: Responsible for the design and construction of the three-axis moving system along with the operation. |
||||||
| 12 | Onboard Edge Computing for High-Resolution FMCW SAR on An Integrated UAV Platform |
Chenxiao Wang Giselle Jeay Jee Lim Victoria Jeay Jia Lim Yinfei Ma |
design_document1.pdf design_document2.pdf design_document3.pdf final_paper1.pdf final_paper2.pdf final_paper3.docx |
Shurun Tan | ||
| # Onboard Edge Computing for High-Resolution FMCW SAR on An Integrated UAV Platform ## 1. Problem Traditional small-scale UAV-borne Synthetic Aperture Radar (SAR) systems suffer from a "blind" data collection process. Because current onboard microcontrollers lack the processing power for complex SAR algorithms, high-resolution 2D images can only be generated via offline processing on a ground station PC after the drone lands. This delay prevents real-time decision-making and limits the immediate usefulness of the UAV in time-sensitive tasks like remote sensing, disaster response, or environmental monitoring. ## 2. Solution Overview Our solution is to develop an integrated real-time imaging system capable of performing edge computing directly on the UAV. We will replace the existing low-performance computing unit with a high-performance embedded edge platform. This allows us to migrate the heavy SAR imaging algorithms from the ground station to the drone itself, converting raw 1D radar waveforms into a 2D top-down terrain map in real-time and providing the operator with immediate visual feedback via a live video stream. As an optional enhancement, we may upgrade the RF frontend by integrating a compact, high-frequency antenna array, which significantly improves scanning resolution while maintaining aerodynamic stability and weight constraints. ## 3. Solution Components ### Onboard Edge Computing Subsystem - High-performance embedded computing platform (e.g., NVIDIA Jetson or equivalent) to replace the legacy low-performance unit (e.g., Raspberry Pi). - Power management circuit to safely draw and regulate power from the UAV battery. ### Software & Transmission Subsystem - Optimized real-time SAR imaging algorithm deployed on the edge computing platform. - Video transmission program to stream the processed 2D map to the ground controller via the drone's API. ### RF Frontend Subsystem (Optional Enhancement) - Compact, high-frequency antenna array for transmitting and receiving microwave signals. - FMCW radar transceiver and Analog-to-Digital Converter (ADC) for raw data acquisition. ## 4. Criterion for Success - The onboard embedded platform must successfully process the raw radar data into a 2D top-down terrain map in real-time (at least 1 frame per second) without exceeding the UAV payload's power limits. - The system must transmit the generated 2D SAR imagery to the operator's remote controller as a live video stream with latency less than 2 seconds, displaying clear structural features rather than abstract 1D waveforms. - If implemented, the upgraded RF frontend and antenna array must successfully capture FMCW backscatter signals during flight while maintaining reduced physical weight to ensure the UAV's aerodynamic stability. TA: Kaiqi Chen |
||||||
| 13 | Autonomous Lawn Patrol Robot for Stray Cat Deterrence |
Chentao Fang Jiawei Kong Ronglong Liu Yanchen Liu |
design_document1.pdf final_paper1.pdf final_paper2.pdf proposal1.pdf |
Yu Lin | ||
| # Problem In many residential neighborhoods, especially in suburban areas in the United States, houses are often surrounded by open lawn spaces. Stray or feral cats may frequently enter these private areas, which can lead to hygiene concerns, property maintenance issues, and disturbance to residents. Existing solutions for preventing animals from entering private yards mainly rely on manual intervention, physical barriers, or simple deterrent devices. These approaches are often inconvenient, inconsistent, or ineffective over long-term use. Therefore, there is a need for a more automated and intelligent system that can monitor outdoor spaces and safely deter unwanted animals. With the advancement of mobile robotics and vision-based sensing technologies, an autonomous patrol robot provides a promising approach to continuously monitor the environment and respond to detected targets. # Solution Overview The proposed solution is an autonomous mobile robot designed to patrol residential lawn areas and deter stray cats. The robot will move within a predefined region using a four-wheel differential-drive chassis. A camera mounted on a servo-driven gimbal will be used to detect and track cat targets through a vision-based recognition system. When a target is detected and confirmed within an appropriate range and direction, a spray-based deterrence mechanism will be activated to safely drive the animal away. The robot’s behavior will be coordinated using a finite state machine that manages transitions between patrol, target tracking, and deterrence modes. Although the intended application scenario is outdoor lawn monitoring, prototype testing and functional validation will be conducted in controlled indoor environments such as tabletop setups due to practical constraints. # Solution Components ## Subsystem I: Mobile Platform - Hardware I.a: Four-wheel differential-drive chassis - Hardware I.b: DC drive motors and wheel assemblies - Hardware I.c: Motor driver module - Software I.d: Basic motion control algorithm ## Subsystem II: Vision and Tracking System - Hardware II.a: Camera module - Hardware II.b: Servo-driven camera gimbal - Software II.c: Cat detection algorithm - Software II.d: Target tracking and alignment logic ## Subsystem III: Deterrence Mechanism - Hardware III.a: Spray nozzle or water outlet - Hardware III.b: Water pump module - Hardware III.c: Turret or nozzle positioning servo - Software III.d: Deterrence activation control ## Subsystem IV: Control System - Hardware IV.a: Microcontroller unit (e.g., Arduino) - Software IV.b: Finite state machine for behavior coordination - Software IV.c: Motor and servo control routines ## Subsystem V: Power and Electronics - Hardware V.a: Rechargeable battery pack - Hardware V.b: Voltage regulation modules - Hardware V.c: Electrical wiring platform (breadboard, perfboard, or PCB) # Criterion for Success - The robot can patrol a predefined test area autonomously or semi-autonomously. - The vision system can detect a cat-like target under typical indoor lighting conditions. - The camera gimbal can maintain stable tracking of the target. - The deterrence mechanism activates only when the target is within a defined range and direction. - The control system correctly switches between patrol, tracking, and deterrence modes. - The differential-drive chassis can perform forward motion and turning maneuvers reliably. - The integrated system demonstrates repeatable performance during multiple indoor validation tests. |
||||||
| 14 | Design of Automated Guided Vehicle Wireless Charging System Based on DSP Position Adaptive Variable Frequency Control |
Jiaxin Cao Jingzhou Ding Jinru Cai Yaxin Li |
design_document1.pdf final_paper1.pdf final_paper2.pdf other1.pdf |
Chushan Li | ||
| # **People** Jingzhou Ding, Jinru Cai, Jiaxin Cao, Yaxin Li # **Problem** In light of national carbon peaking and carbon neutrality goals, smart electric vehicles are a backbone force in reducing carbon emission. However, one of the largest obstacles in the promotion of EVs is the capacity of batteries. Wireless Power Transfer (WPT) offers a promising solution, but efficiency has long been a problem due to low mutual induction and a low transmission factor between the transmitter and receiver coils. Furthermore, this inefficiency is often exacerbated by foreign metal objects and the misalignment of the vehicle while parking at the charging station. # **Solution Overview** We propose an integrated prototype of a wireless charging system consisting of a mobile robot (car) and a charging station. To solve the misalignment and efficiency issues, the system integrates computer vision to achieve environmental perception and precise position detection. Instead of relying solely on physical alignment, the system uses the visually calculated position offset to feed a DSP controller. The DSP dynamically adjusts the switching frequency of a Dual Active Bridge (DAB) resonant converter. By combining this position-adaptive frequency control with a hill-climbing algorithm for Maximum Power Point Tracking (MPPT), the circuits of the transmitter and receiver sides are continuously tuned to maintain resonance and maximize power transfer efficiency. # **Solution Components** ## Visual and Navigation Subsystem (ECE) - Raspberry Pi and Camera Module for integrating computer vision technology to detect the charging station and calculate spatial offset. - STM32 Microcontroller for autonomous navigation and obstacle avoidance. ## Power Transfer and Control Subsystem (EE) - Tx and Rx Coils arranged at the bottom of the car to save space. - DSP controller for executing the MPPT algorithm and providing variable switching frequencies. - Dual-Bridge Series Resonant Converter (DBSRC) circuit, including an inverter network, high-frequency transformer, and rectifier network. ## **Criterion for Success** - The car must successfully and automatically detect the location of the wireless charging stations and navigate to the charging point. - The visual tracking system must accurately output the relative position coordinates of the coils to the DSP. - The WPT system must support a wireless fast charging capability of ≥ 20W to achieve efficient energy replenishment. - The DBSRC and DSP control logic must successfully impose zero-current switching or zero-voltage switching under load or positional variations. |
||||||
| 15 | Vision-Based Sign Language Recognition System for Smart Furniture Control |
Chongying Yue Licheng Xu Mingzhi Gu Zihan Xu |
design_document1.pdf final_paper1.pdf final_paper2.pdf final_paper3.pdf other1.pdf proposal1.pdf |
Yushi Cheng | ||
| ## Problem Current smart home systems rely primarily on voice control or mobile apps for operation. However, these interaction methods are not user-friendly for the hearing impaired, and controlling furniture devices via mobile apps requires additional steps, resulting in low interaction efficiency. Therefore, this project aims to develop a system that can directly control furniture devices through visual gesture recognition, providing a more intuitive and accessible interaction method for smart homes. ## Solution Overview Our solution is a vision-based sign language recognition smart furniture control system. The system uses a camera to capture the user's hand movements in real time and utilizes computer vision technology to detect key hand points and gestures, converting them into corresponding furniture control commands, *such as turning on the lights*. The system sends the gesture recognition results to the main control unit, where the main controller parses the control commands and generates corresponding control signals to drive the furniture devices. ## Solution Components ### Software Component - **Real-time Gesture Recognition**: real-time gesture recognition on the vision processing unit. The system acquires hand images through a camera and uses MediaPipe to extract gesture features. Based on these features, a lightweight machine learning model classifies gestures and recognizes the user's input control gestures. - **Control Logic**: The main controller receives gesture recognition results from the vision recognition module and parses them into specific control commands. The system generates PWM or GPIO control signals based on different commands to drive physical devices. ### Hardware Component - **Vision Processing Unit**: Includes a camera module and vision processing board *(e.g., K230)* , which acquires user hand images and running gesture recognition algorithms. - **Main Control Unit**: An STM32 microcontroller used to receive recognition results and generate corresponding control signals. - **Execution Drive Module**: Motor drive circuits and relay modules control the actual furniture devices, *e.g., smart lighting systems*. ## Criteria of Success - The system can stably recognize at least 5 predefined gestures with an accuracy rate of over 70%. - The system latency from user gesture input to furniture device response is less than 1 second. - The system can successfully control at least two types of furniture devices. ## Distribution of Work - **Zihan Xu** Develops the visual recognition module and is responsible for testing the accuracy of gesture recognition under different environments. - **Licheng Xu:** Designs STM32 control programs, parsing gesture commands, and generating PWM/GPIO control signals. - **Chongying Yue:** Responsible for hardware circuit design and implementation, including motor drive circuits and power management. - **Mingzhi Gu:** Responsible for system architecture design and overall integration, including the design and debugging of the furniture control interface and system stability testing. |
||||||
| 16 | Design of a Raspberry Pi–based monitoring system for shared living environments |
Denghan Xiong Jihao Li Mujia Li Shixuan Ma |
design_document1.pdf final_paper1.pdf proposal1.pdf |
Chao Qian | ||
| # Problem In shared living environments such as dormitories or shared apartments, traditional access control methods such as keys or passwords can be inconvenient and insecure. Keys can be lost or copied, and passwords can be shared or forgotten. These problems make it difficult to ensure security and manage access effectively. In addition, traditional door locks cannot automatically identify who is entering or keep records of access events. Residents or administrators may want a system that can automatically recognize authorized users and log entry activities for security purposes. Therefore, there is a need for a lightweight and low-cost smart access control system that can detect people approaching the door, identify authorized users, and manage access automatically. # Solution Overview Our solution is to design a smart access control system based on a Raspberry Pi for shared living environments such as dormitories or shared apartments. The system uses a PIR motion sensor to detect human presence near the door. When motion is detected, a USB camera connected to the Raspberry Pi captures images of the person standing at the door. The captured images are processed using computer vision techniques, and a face recognition algorithm is used to determine whether the person is an authorized user. If the user is recognized, the system activates a relay module to simulate unlocking the door. If the person is not recognized, the system records the event and displays the result. The system also includes a graphical user interface developed with PyQt5 to display the camera feed, recognition results, and system status. A local database is used to store user information and access records. # Solution Components ## Subsystem I – Hardware System ### Hardware I.a Motion Detection Sensor - A PIR motion sensor detects human movement near the door. - When motion is detected, the sensor sends a signal to the Raspberry Pi. - This signal triggers the image capture and recognition process. ### Hardware I.b Camera Module - A USB camera connected to the Raspberry Pi captures images of the person at the door. - The camera provides real-time video frames for face detection and recognition. ### Hardware I.c Door Control Module - A relay module controlled by the Raspberry Pi simulates the door unlocking mechanism. - When an authorized user is detected, the relay activates to unlock the door. ## Subsystem II – Image Processing and Recognition ### Software II.a Image Capture and Processing - OpenCV is used to capture video frames from the camera. - Images are preprocessed using techniques such as resizing and color conversion. ### Software II.b Face Recognition Module - A face recognition algorithm extracts facial features from captured images. - These features are compared with stored user data to determine the person's identity. ## Subsystem III – User Interface and System Management ### Software III.a Graphical User Interface - A graphical interface built with PyQt5 displays the camera feed. - It shows recognition results and system status in real time. ### Software III.b Multithreading Framework - A multithreaded architecture allows the system to perform multiple tasks simultaneously. - Tasks such as video capture, face recognition, and interface updates run in parallel. ### Software III.c Local Database Management - A local database stores authorized user information. - The database records access logs such as time, identity, and recognition results. # Criteria of Success - The PIR sensor can detect human motion and trigger the system automatically. - The camera can capture images successfully for processing. - The face recognition module can correctly identify authorized users from stored data. - The relay module activates to simulate unlocking the door when a valid user is detected. - The graphical interface displays the camera feed and recognition results in real time. - The system records access events in the local database. - The system runs smoothly on the Raspberry Pi with minimal delay. |
||||||
| 17 | Machine Vision-Based Intelligent Fruit and Vegetable Picking & Sorting Robotic Arm |
Fengyi Jin Shengyu Xu Simeng Yan Wenye Zhang |
design_document1.pdf final_paper1.pdf final_paper2.pdf final_paper3.docx |
|||
| # Problem In agricultural production, the classification of produce (e.g., fruits and vegetables) based on dimensions or chromatic features is frequently required. Manual sorting is prone to error and incurs significant labor costs; conversely, a robotic arm specifically engineered for this task enables continuous 24-hour operation while substantially reducing operational expenses. # Solution Overview The robotic system leverages computer vision to recognize the size and color of workpieces, facilitating the real-time transfer of spatial coordinates and attribute data to the controller. The control architecture then drives the actuators to align the flexible end-effector with the target for autonomous grasping, followed by precise sorting into predefined areas based on the detected classifications. # Conponents ## Robotic Arm Composed of 4-6 motors and rigid arm. Able to move the end effector to specific position ## Gripper Penumatic Soft gripper with two fingers to grip objects without harm ## Machine Vision System Recognize object, category and send object position ## Control unit Get object position from vision system. Control robotic arm to certain position. Control gripper to grip. # Criteria of Success Be able to category according to color or size. Be able to grasp and put objects to certain areas according to their categories. |
||||||
| 18 | Real-Time Traffic Monitoring and Congestion Analysis Using Raspberry Pi and Computer Vision |
Ding Jiang Yiyang Cheng Yucong Gao Zetong Lang |
design_document1.pdf final_paper1.pdf final_paper3.pdf |
Chao Qian | ||
| # Problem Monitoring urban traffic congestion is a growing challenge for city infrastructure management. Current monitoring methods often rely on manual observation or expensive dedicated hardware, and lacking time sensitivity, making real-time traffic density analysis difficult and costly. Traffic managers lack accessible, affordable tools to continuously monitor vehicle counts and congestion levels, limiting their ability to optimize traffic flow efficiently. # Solution overview Our solution is a Raspberry Pi-based traffic monitoring system that uses a camera to capture and analyze traffic density in real-time. This system applies image processing to count vehicles and monitor congestion levels, providing data for optimizing traffic flow. We will also include a live dashboard to visualize traffic density data and to generate congestion alerts automatically, providing an affordable and scalable tool for traffic flow optimization. # Solution component ## Software component - Real-time vehicle detection and counting module implemented on Raspberry Pi using computer vision techniques (e.g., OpenCV or lightweight deep learning models). This module captures video streams from the camera and processes frames to identify and count vehicles in traffic. - Traffic data processing and visualization module that collects vehicle count data, estimates traffic density, and sends the processed data to a real-time dashboard. The dashboard will visualize traffic conditions and generate congestion alerts when vehicle density exceeds predefined thresholds. ## Hardware component - Raspberry Pi computing unit integrated with a high‑resolution camera module to capture continuous traffic video streams and perform edge computing for image processing. - Weather-resistant outdoor enclosure with an adjustable camera mounting structure. The enclosure protects the electronics from rain, dust, and temperature variations while maintaining stable camera positioning for reliable long-term monitoring. # Criteria of success - The dashboard should collect proper amount of data to determine the extent of congestion and to send alarms at certain level of congestion. - The whole structure should work continuously under bad weather condition such as rain, snow, etc. - The dashboard should display relevant traffic data with as minimal latency as possible and report timely. # Distribution of work ## Ding Jiang: Responsible for developing the core image processing algorithms on the Raspberry Pi. This includes implementing vehicle detection and counting using computer vision techniques, optimizing the processing pipeline for real-time performance, and integrating the camera input with the processing module. This member also assists in testing the accuracy of vehicle detection under different traffic conditions. ## Yucong Gao: Responsible for building the real-time traffic monitoring dashboard. Tasks include designing the interface for visualizing vehicle counts and congestion levels, implementing the data pipeline from the Raspberry Pi to the dashboard, and developing congestion alert functions. This member also ensures that the system provides low-latency updates and reliable visualization of traffic density data. ## Yiyang Cheng: Responsible for the hardware setup of the Raspberry Pi traffic monitoring system. This includes installing and configuring the Raspberry Pi, camera module, and power supply system. This member also handles hardware integration and ensures that the camera and computing unit operate reliably for continuous video capture. ## Zetong Lang: Responsible for designing the outdoor enclosure and camera mounting structure. The enclosure must protect the system from environmental conditions such as rain, dust, and temperature changes while allowing stable camera positioning. This member will design and prototype the enclosure, ensuring durability and ease of installation. |
||||||
| 19 | Vision-Guided Sorting and Pickup Cleaning Robot |
Dailin Wu Jinyang Chen Tinghao Pan Zihan Zhou |
design_document1.pdf design_document2.pdf final_paper1.pdf final_paper2.pdf final_paper3.pdf other1.mp4 other2.pdf |
Meng Zhang | ||
| # Vision-Guided Sorting and Pickup Cleaning Robot ## 1. Problem Definition and Motivation Public environments such as campuses, parks, and sidewalks often accumulate scattered trash that requires frequent manual cleaning. Traditional cleaning methods rely heavily on human labor, which can be inefficient and costly for large or continuously used spaces. In addition, many existing robotic cleaning systems mainly focus on navigation and simple sweeping functions but lack the ability to intelligently identify and sort waste. To address this limitation, this project aims to develop a vision-guided autonomous cleaning robot capable of detecting, classifying, and collecting trash objects. By combining computer vision with robotic manipulation, the robot can not only identify waste items but also physically remove them from the environment. This integrated perception-to-action pipeline allows the system to perform both cleaning and basic waste sorting automatically. The success of this project will be evaluated based on the following criteria: - The robot can reliably detect and identify trash objects in its field of view. - The system can classify waste into predefined categories. - The robotic arm can successfully pick up and relocate detected waste items. - The system can operate autonomously with minimal human intervention. --- ## 2. Solution Overview The proposed solution integrates vision-based perception and robotic manipulation into a unified workflow. An onboard camera captures images of the surrounding environment, and a computer vision model analyzes these images to locate potential objects and determine whether they should be treated as waste. Once an object is identified as garbage, the system assigns it to a waste category. This classification is not only used for sorting but also helps guide the manipulation strategy. Different waste categories may correspond to different object shapes, sizes, or surface properties, which influence how the robotic arm approaches and grasps the item. Compared with conventional cleaning robots that only sweep debris or rely on predefined object shapes, the proposed system introduces visual intelligence and adaptive grasping, enabling the robot to handle a wider variety of waste items. The feasibility of the system is supported by the availability of common hardware components such as cameras, embedded processors, and robotic arms, as well as existing computer vision models that can be adapted for object detection and classification. --- ## 3. System Architecture and Components ### Vision Module The vision module captures images using an onboard camera and processes them through a trained vision model to detect objects and classify potential waste. The output of this module includes the object’s location, category, and estimated properties. ### Decision and Planning Module Based on the detection results, this module determines whether the object should be collected and calculates the appropriate grasping strategy. It generates the required motion commands for the robotic arm. ### Manipulation Module The robotic arm performs the physical pick-and-place action. The arm adjusts its approach direction, grasp point, and gripping force to accommodate different types of waste objects. ### Sorting and Storage Module After an object is successfully grasped, it is placed into the corresponding container or storage area according to its waste category. |
||||||
| 20 | Magnetic-Wheeled Pipe Climbing Robot for PowerTool Rust Removal |
Huanyu Feng Junxiang Qin Xiaocheng Zhang Xuhao Yang |
design_document1.pdf final_paper1.pdf proposal1.pdf |
Jiahuan Cui | ||
| # Problem Corrosion of steel infrastructure causes trillions of dollars in economic losses globally each year. For external steel pipework, rust removal is a necessary prerequisite for reliable coating adhesion. However, traditional manual grinding and wire brushing are labor-intensive and expose workers to hazards such as working at heights, confined spaces, dust, and vibration. Furthermore, manual cleaning often results in inconsistent surface quality, which negatively impacts long-term corrosion resistance. # Solution Overview The proposed solution is a magnetic-wheeled pipe-climbing robot capable of automated power-tool rust removal on external steel pipes. By automating the grinding and brushing process, the robot reduces human exposure to extreme and hazardous maintenance conditions. To ensure consistent surface quality, the robot features a flexible end effector with force-position adjustment to stabilize tool-to-surface contact over pipe curvatures and irregularities. # Solution Components ## Mobility and Adhesion Subsystem • Compact 4-wheel chassis compatible with the curvature of steel pipes. • Drive motor assembly for locomotion. • Magnetic adhesion wheels designed to provide passive, continuous attraction and sufficient traction margin to prevent slippage during grinding. ## Rust Removal End Effector Subsystem • Modular, interchangeable abrasive tool head to mount attachments like wire brushes, fiber discs, or nonwoven conditioning discs. • Tool head motor with controllable rotational speed. • A flexible normal mechanism, such as springs or a leadscrew motor, to provide mechanical buffering and absorb vibration. ## Control and Sensing Subsystem • Force sensor or load cell to measure normal contact force. • Closed-loop control system that uses force sensor data and the leadscrew motor to implement a normal position compensation loop. • Protection logic system to handle abnormal events like tool jamming, sudden loss of adhesion, or emergency stops. # Criteria of Success • The robot must maintain reliable magnetic adhesion and traction on curved steel surfaces without slipping while under tangential tool loads. • The rust removal end effector must successfully maintain stable normal contact force when encountering surface irregularities like welds, pits, and thickness variations. • The system must achieve specific target cleanliness grades (e.g., ISO 8501-1 St 2/St 3 or SSPC-SP 11) and surface roughness/profile metrics required for industrial coating preparation. |
||||||
| 21 | Vision-driven Automatic Posture Correction Device |
Weichong Chen Xiaoyu Xu Yilun Chen |
design_document1.pdf final_paper1.pdf final_paper2.pdf presentation1.pptx proposal1.pdf |
Wee-Liat Ong | ||
| #Problem The digital age has led to increased reliance on portable electronic devices, causing a significant rise in poor sitting postures. Traditional brackets lack dynamic adjustment, forcing users to adapt to fixed screens, which hinders healthy habits. Existing market solutions often fail to optimize the sight-screen relationship or rely on imprecise manual adjustments. This results in health issues such as cervical spine strain, muscle soreness, and carpal tunnel syndrome. #Solution Overview The project develops an Automatic Sight Correction Device Bracket. Using visual detection and attitude sensing, the bracket dynamically adjusts its height and tilt angle to maintain the user’s sight in a horizontal state. It is portable, universally compatible with mainstream devices, and powered via USB for mobile use. #Solution Components ##Subsystem 1 (Hardware) Core Microcontroller: A high-performance MCU processes data from the camera and gyroscope to perform closed-loop regulation of actuators. Actuators: Includes a micro electric linear actuator (load capacity ≥2kg) with a linear encoder for precise height adjustment, and a worm gear motor for tilt angle control. Sensing Modules: A high-definition camera captures facial landmarks (pupil, jawline) with autofocus and low-light compensation. A gyroscope provides real-time attitude data. Structure & Power: Built from Higher-strength 3D printing materials, the frame supports 7-12.9 inch tablets, smartphones and e-readers. It uses a 5V/2A USB-C power scheme. Interaction: Includes a one-click start button, emergency stop, and a DIP switch for manual/automatic mode switching. ##Subsystem 2 (Software) Data Processing: Uses the Kalman Filter Algorithm to fuse sensor data and the Mediapipe framework to detect 68 facial landmarks. Control Logic: A PID Control Algorithm calculates deviations between actual sight and the horizontal standard, driving actuators to correct the bracket without overshoot. Safety: Automatically triggers alarms and stops adjustment if feedback is lost or deviations persist. #Criteria of Success Efficiency: Completes initial correction within 10 seconds; responds to posture changes exceeding 3°. Performance: Supports up to 2kg loads; achieves ≥95% recognition accuracy under various lighting. Safety & Usability: Features torque protection and emergency stop. Setup involves only three steps: place device, fix bracket, and one-click start. |
||||||
| 22 | Smart Climate-Controlled Rice Dispenser |
Gaoning Zhao Shining Wang Yuyang Liu Zixin Yang |
design_document1.pdf design_document2.pdf final_paper1.pdf presentation1.pptx video1.mp4 |
Bruce Xinbo Yu | ||
| # Problem Rice is a staple food sensitive to environmental factors; improper storage in high temperature or humidity leads to mold, pest infestation, and loss of nutritional value. Additionally, traditional manual dispensing is often imprecise, resulting in food waste. Current market solutions are largely passive containers that fail to monitor or actively preserve rice quality. # Solution Overview We propose an integrated smart dispenser that provides active climate regulation and precise, automated dispensing. The system will monitor internal conditions and utilize a control loop to maintain an ideal storage environment, while a motor-driven mechanism ensures accurate portioning based on user input. # Uniqueness & Innovation This project is an innovation that bridges the gap between traditional food storage and automated home appliances. While competitors offer airtight seals, our system introduces active thermal management and sensor-driven feedback to guarantee freshness. # Technical Overview The system consists of three main sub-modules: - Climate Control: A sensor network (Temperature/Humidity) feeding into a microcontroller to actuate a thermal management structure (fans/dehumidifiers). - Precision Dispensing: The initial UI will feature tactile buttons for preset quantities (e.g., 100g, 200g), with a modular design that allows for upgrading to a touchscreen for custom numeric input depending on the implementation schedule. - UI that displays real-time environment status. # Criterion for Success Successfully maintain storage humidity below 60% and temperature below 25°C (or 5°C below ambient). Achieve dispensing accuracy within ±5% of the user-defined weight. Functional UI that triggers the correct dispensing weight and displays real-time environmental status. |
||||||
| 23 | Boost Converter Design Agent based on PE-GPT and PANN |
Haojun Li Jiaming Ma Jiarong Xu Xinyu Zhan |
design_document1.pdf final_paper1.pdf presentation1.pptx presentation2.pptx |
Fanfan Lin | ||
| ## **People** Jiarong Xu, Haojun Li, Xinyu Zhan, Jiaming Ma ## **Problem** Designing power electronics, such as boost converters, traditionally requires bridging a significant gap between high-level natural language design requirements and complex physics-based hardware realization. This design process is often time-consuming, requires extensive domain expertise, and relies on disjointed tools for calculation, simulation, and troubleshooting. Currently, the industry lacks an intelligent, automated closed-loop workflow. There is a critical need for an agent, which can autonomously plan and decompose the design workflow for users, seamlessly translating high-level requirements into verified component parameters, guiding physical assembly, and executing automated hardware diagnostics. ## **Solution Overview** We propose a "Boost Converter Design Agent" based on PE-GPT and PANN to achieve an automated, closed-loop workflow. Central to this system is PE-GPT's ability to act as an intelligent assistant that autonomously plans and decomposes the entire complex hardware design workflow for the user. The system follows a "Brain-Planning-Tool" architecture: The LLM "Brain" (GPT-4 with RAG) handles natural language understanding. The "Planning" module uses Chain-of-Thought (CoT) reasoning to translate initial user requirements into a clear, step-by-step execution plan (spanning Planning, Design, Assembly, and Diagnosis phases), guiding the user through the process. Subsequently, the agent invokes the "Tool" module (integrating PANN) for fast forward-mode simulation (generating dynamic waveforms to validate the design) and inverse-mode diagnosis. Overall, the agent does not merely calculate optimal L and C parameters; it orchestrates the entire process, guiding users step-by-step to assemble PCB modules and analyzing experimental data to complete a closed-loop diagnosis. ## **Solution Components** **1. AI Agent and Interactive Planning Subsystem (Software)** * LLM Core (GPT-4) & RAG Module: Responsible for natural language understanding and retrieving specialized domain knowledge * for power electronics. * Workflow Planner (CoT Reasoning Controller): The core of the planning process. It interacts with the user to autonomously decompose high-level design tasks into a strictly logical, step-by-step plan before execution, providing the user with a clear global view and operational steps. **2. Simulation and Diagnostic Subsystem (Software/Algorithm)** * PANN Forward Mode (Fast Simulation): Acts as a fast simulation tool to generate dynamic waveforms based on the agent's determined L and C parameters, validating the design prior to physical assembly. * PANN Inverse Mode (Twin Diagnosis): Analyzes experimental data gathered from the physical circuit, diagnoses performance deviations, and provides specific adjustment suggestions. **3. Hardware Execution and Data Acquisition Subsystem (Hardware)** * Plug-and-Play PCB Modules: Modular inductors, capacitors, and switching components that allow users to quickly configure and assemble the physical boost converter, guided by the steps planned by PE-GPT. * Data Acquisition Unit (Sensors/Microcontroller): Used to capture real-time experimental data (voltage/current waveforms) from the assembled boost converter and feed it back to the diagnostic system for closed-loop analysis. ## **Criterion for Success** * Successful Workflow Planning: The PE-GPT agent must successfully interpret a natural language design request and output a clear, logical, step-by-step design workflow plan for the user. * Successful Parameter Design: The agent must accurately calculate the optimal component parameters (Inductance L, Capacitance C) for the boost converter. * Successful Simulation Validation: The PANN tool must successfully generate dynamic waveforms that validate the feasibility of the hardware design prior to physical assembly. * Successful Hardware Assembly: Guided by the agent's planned steps, the user must successfully assemble the physical boost converter using the plug-and-play PCB modules and achieve the intended voltage boost function. * Successful Closed-Loop Diagnosis: The Twin Diagnosis module must successfully ingest physical experimental data, identify discrepancies between the physical hardware and theoretical simulation, and output valid adjustment suggestions. |
||||||
| 24 | StepWise: A Smart Insole System for Real-Time Gait Analysis and Muscle Rehabilitation |
Kerui Xie Nuo Pang Xiaorui Zhang Zhichao Chen |
design_document1.pdf final_paper1.pdf final_paper2.pdf other1.docx proposal1.pdf |
|||
| # Problem Despite walking being the most fundamental human movement, the majority of the population suffers from undiagnosed gait abnormalities such as Pes Planus (Flat Feet), over-pronation, and irregular strike patterns, which lead to a "kinetic chain" of health failures. When a person suffers from improper heel-to-toe transition, the misalignment doesn't stay in the foot. It forces the ankles to roll inward or outward, the knees to rotate unnaturally, and the pelvis to tilt. This creates a "Kinetic Chain" reaction that is a primary driver for chronic lower back pain, hip bursitis, and premature osteoarthritis in the knees. Because these issues develop slowly over years, most individuals do not realize their walking posture is the root cause until permanent joint damage has occurred. Currently, high-fidelity gait analysis is confined to specialized medical facilities. Systems like optical motion capture (Vicon) or pressure-sensitive walkways (GAITRite) cost tens of thousands of dollars and require trained clinicians to operate. Existing solutions are mostly passive. Traditional orthotic inserts act as a "crutch" for the foot, supporting the arch without actually strengthening the muscles responsible for maintaining that arch. # Solution Overview StepWise is a 3D-printed smart insole that replaces expensive clinical gait labs with a wearable diagnostic ecosystem. It features a 5-vital-point pressure sensor array to capture real-time foot mechanics and orientation during daily activities. Data is transmitted via Bluetooth to a mobile app, where algorithms identify pathologies like Flat Feet (Pes Planus) and muscle fatigue. To close the loop, the system provides haptic alerts for posture correction and recommends personalized exercises to actively strengthen the user's foot muscles and prevent chronic joint pain. # Solution Components ## Data Collection System - Pressure sensor array for capturing high-resolution pressure data at five anatomical foot points - Sensors for tracking foot orientation, swing velocity, and strike angles - Circuits to convert resistance change to voltage values ## Transmission System - Low-power data processing and wireless Bluetooth transmission - LiPo power management unit and protection circuit ## Diagnostic and Feedback System - Mobile Application for real-time visualization of foot pressure heat maps - Analysis Algorithm to identify pathologies - Provide physical alerts to correct poor walking posture # Criterion for Success - Successful capture of distinct pressure signatures from all five sensor points with high resolution - The 3D-printed insole must withstand 500 compression cycles - Collection system lasts for 1 hour - The algorithm accurately distinguishes between "Heel Strike," "Mid-Stance," and "Toe-Off" - Successful detection of pathologies - Generate personalized exercise routine based on the user's specific muscle fatigue data |
||||||
| 25 | AR-based Palm-size Robotic Assistant 1 |
Fengwei Yang Jiaqi Ding Ruixi Qin Yuzhang Wang |
design_document1.pdf design_document2.pdf final_paper1.pdf other1.pdf |
Liangjing Yang | ||
| # Problem Portable robotic assistants are increasingly integrated into daily activities. However, a significant bottleneck remains: the control interfaces are often unintuitive and user-unfriendly. Most existing systems rely on abstract buttons or 2D joysticks that do not provide spatial context, leading to a steep learning curve and inefficient operation. There is a critical need for a control method that allows users to interact with robots in a more natural, spatial, and visual manner # Solution Overview This project aims to design and implement an Augmented Reality (AR) based smartphone-controlled robotic assistant. The project will replace traditional control schemes with an intuitive AR interface. This system will allow users to issue commands directly through the smartphone's camera view, creating a seamless bridge between human intent and robotic execution. # Solution Components ## Subsystem 1 (Hardware) Palm-sized Robot Chassis: A compact mobile base equipped with micro-motors and basic obstacle sensing. Control Unit & Communication: An onboard microcontroller (e.g., ESP32) to handle real-time movement commands received via Wi-Fi/Bluetooth. ## Subsystem 2 (Software) AR Visualization Layer: An app that overlays digital UI (path lines, status markers) onto the live robot feed. Spatial Mapping & Interaction: Implementing "Point-to-Move" functionality where the user taps a location in the AR view to send coordinates to the robot. # Criteria of Success The project will be successful if the AR interface demonstrates a measurable improvement in interaction intuitiveness. Specifically: The user can guide the robot to a target location by interacting solely with the AR smartphone interface. The system maintains stable spatial alignment between the virtual UI and the physical robot. The control latency is low enough to ensure real-time responsiveness. |
||||||
| 26 | An Engineering Solution to a Local Indoor Safety and Comfort Regulation System |
Kaicheng Wu Wenyu Zhou Yichen Zhang |
design_document1.pdf final_paper1.pdf final_paper2.pdf final_paper3.pdf presentation1.pptx |
Meng Zhang | ||
| #Problem Modern homes require both comfort and safety, but many indoor environmental changes still need to be handled manually. A sudden drop in temperature or low humidity can reduce comfort and affect health, while smoke, gas leakage, or early fire conditions can quickly become dangerous if they are not detected in time. Existing household devices often work independently and lack coordinated automatic response. Some smart home systems also rely on network connectivity or pre-built software platforms, which may reduce reliability when fast local action is needed. #Solution Overview We will design a compact embedded hardware system that monitors key indoor environmental conditions and responds automatically through local control. The system will be based on an STM32 microcontroller and will focus on real-time sensing, local decision making, and coordinated actuation without relying on existing smart home software platforms. Our design will monitor temperature, humidity, smoke, and gas-related hazards. Based on these conditions, the system will regulate the environment or issue warnings through hardware responses such as ventilation, heating, humidification, and alarms. The final prototype will prioritize functional integration, compact size, and reliable demonstration rather than furniture-like appearance. #Solution Components Environmental Sensing Module This module measures temperature, humidity, smoke concentration, and combustible gas indicators. It provides real-time environmental data for system monitoring and hazard detection. #Embedded Control Module This module uses an STM32 microcontroller as the core of the system. It collects sensor data, determines system state, and coordinates local responses based on predefined control logic. #Response and Actuation Module This module performs the physical response of the system. It can activate devices such as a ventilation fan, heating element, humidification unit, buzzer, or warning lights according to the detected condition. #User Interface Module This module displays system status and supports simple user interaction. It may include a small display, indicator lights, and control buttons for observing readings and warning states during demonstration. #Criteria of Success Our product should be able to continuously monitor major indoor environmental conditions and detect abnormal situations in real time. The system should automatically trigger the proper local hardware response when temperature, humidity, smoke, or gas conditions move beyond acceptable ranges. In hazardous situations, the prototype should prioritize warning and protective actions such as alarms and ventilation. The final system should operate as a compact and integrated hardware prototype with coordinated sensing, control, response, and user feedback. |
||||||
| 27 | Smart Foot-Controlled Mouse with Sensor Fusion and UI-Aware Assistance |
Chaoxiang Yang Hao Liu Jiongye Liu Zhihao Cheng |
appendix1.docx design_document1.pdf design_document2.pdf final_paper1.pdf final_paper2.pdf presentation1.pptx video1.mp4 |
Wee-Liat Ong | ||
| Team Members Zhihao Cheng(zhihao10) Hao Liu(hao25) Chaoxiang Yang(cy60) Jiongye Liu(jl244) #Problem In the present and in the foreseeable future, the interaction of electronic devices still heavily relies on devices such as keyboards and mice. However, these devices have overlooked certain groups of people who are unable to use them, such as users with upper limb disabilities. For instance, it is often difficult for them to use ordinary mice that require precise hand control. Due to the widespread use of computers in our daily lives, we hope that people with disabilities can break free from the shackles of certain device limitations. To address the issue of inconvenient interaction for disabled individuals, there are already some assistive devices available on the market. Eye-tracking systems work well, but they are often expensive. Simple foot switches are easy to operate, but they can only handle simple commands. Therefore, we propose manufacturing wearable foot-controlled mice, which will provide users with a more complete and practical way of using computers. This approach may also provide an alternative interaction method for human-computer interfaces. #Solution Overview Our project is to design a smart wearable device similar to a pair of slippers, namely a smart foot-controlled mouse. Firstly, multiple pressure sensors are used to detect foot gestures. These data will be converted into mouse movements and mouse operations (such as clicking, double-clicking, and dragging) through algorithms. If the operation is allowed, it can also support custom key selection. The system also includes auxiliary functions that are compatible with the user interface to assist in achieving precise movement operations. For example, the system can detect small icons or interactive links, and intelligently reduce the mouse movement speed to achieve precise clicks, making it easier to perform delicate tasks. #Solution Components ##Sensing Hardware - FSR pressure sensors for detecting foot pressure distribution - The sensor signals are used to distinguish different foot gestures ##Embedded Control Hardware -Sensor data collection and signal processing -The controller generates signals into mouse commands ##Communication and Power Hardware - The communication hardware sends control signals to the computer. - The system supports either wired or wireless communication with the computer. - A rechargeable battery or wired power supply can be used for power. ##Device Structure - An ergonomic structure designed for comfortable and stable foot interaction. - A lightweight structural frame used to hold sensors and electronic components. - Durable materials are used to improve stability and long-term reliability. ##Software -Foot gesture recognition and signal processing -Cursor movement control and mouse command generation -Customization for sensitivity adjustment -Provide UI-aware assistance for precise interaction #Criterion for Success -The device should move the cursor smoothly and perform basic mouse commands like click, double-click and drag. -The mis-trigger rate should be relatively low in standard testing conditions. -The system should reliably distinguish different foot gestures after a short calibration process. -The wearable structure should stay stable and reasonably comfortable during extended use. -The device should work with common computers using either wired or wireless communications. -The UI-aware assistance should help users select small on-screen targets more easily than basic foot-only control. #Distribution of Work Jiongye Liu is responsible for the Power Hardware and Device Structure, including power support, ergonomic design, and hardware integration. Zhihao Cheng, Hao Liu, and Chaoxiang Yang are responsible for the Sensing Hardware, Communication Hardware, Embedded System, and Software, including signal collection and processing, gesture recognition, cursor control, mouse command generation, sensitivity adjustment, and UI-aware assistance. |
||||||
| 28 | Extended Reality Based Robotic Desktop Assistant |
Cheng Zheng Yuxuan Wu Zhewei Zhang Ziyang Jin |
design_document1.pdf final_paper1.pdf final_paper2.pdf other1.docx proposal1.pdf video1.mp4 |
Liangjing Yang | ||
| #People Cheng Zheng: cz77 Yuxuan Wu: yuxuan59 Ziyang Jin: ziyang3 Zhewei Zhang: zheweiz3 #Problem: Portable robotic assistants have strong potential for everyday use, yet their compact form factor severely limits on-board user interfaces. As a result, users often cannot quickly understand what the robot can do, what it is currently doing, and how to interact with it efficiently. Most palm-size robots rely on a mobile app or a few physical buttons, which leads to a narrow and less intuitive interaction style. Users frequently need to switch between checking the phone, issuing commands, and observing the robot’s response, increasing both the learning curve and operational friction. Some existing solutions enhance interaction by adding external displays or extra devices, but this increases system bulk and setup complexity, undermining portability and “grab-and-go” usability. By turning any flat surface (e.g., a desk or a wall) into an interactive projection-based interface and integrating gesture recognition with dynamic visual feedback, the robot can provide a more natural and direct human–robot interaction experience without requiring an additional screen. This approach improves usability by making robot status and functions easier to perceive and operate, while also enabling an “interface anywhere” form factor that better fits real-world daily-assistance scenarios and enhances user engagement. #Solution Overview: Core function: ##Dynamic Projection Interface: The robot projects an interactive user interface onto any flat surface (e.g., a desk or a wall), converting the surrounding physical space into an operable interaction area without adding an external display. ##Gesture-Based Interaction Control: Users interact with the projected interface using hand gestures. The system performs real-time gesture detection and recognition, maps gestures to commands, and triggers corresponding robot responses, enabling a natural and intuitive interaction flow. ##Interface Navigation: The main projected interface provides basic feature entries (e.g., Weather, Clock, Exit) and supports page switching and function invocation through a “point-and-click” interaction style. ##Information Query Functions: Selecting the Weather icon switches the interface to display current weather information. Selecting the Clock icon switches the interface to display the current time. Each sub-page includes an Exit icon that returns the user to the main interface, ensuring a consistent and easy-to-learn navigation logic. ##Affective (Emotional) Interaction: Simple gesture-triggered feedback is included to improve engagement and user friendliness. A thumbs-up gesture triggers a 👍 animation with a cheerful sound; a thumbs-down gesture triggers a 👎 animation with a sad sound; and a heart gesture triggers a ❤️ animation with a warm, gentle sound. #Components: 1. Mechanical Module The mechanical module is designed to meet the overall goal of a palm-size, mobile robot with a projection-based interactive interface. A compact and lightweight structure is adopted to ensure stable desktop mobility, provide attitude adjustment, and support proper mounting and viewing angles for the projector–camera system. The module consists of three main parts: (1) Omni-wheel Mobile Base: a three-omni-wheel chassis is used, with each wheel diameter no larger than 5 cm, enabling agile planar motion and maneuverability on desktop surfaces; (2) Mini Gimbal: a 2-DoF gimbal provides orientation adjustment with a pitch range of approximately ±30°, allowing the system to align the projection and vision direction under different usage conditions and improving projection/recognition robustness; (3) Lightweight Enclosure: the enclosure will be fabricated via 3D printing to support rapid iteration and assembly optimization. The overall robot size is constrained within 15 cm × 15 cm × 15 cm to maintain portability and desktop friendliness. 2. Electronic Module The electronic module is selected and integrated to support a low-power, portable, and fully self-contained system. It provides computation and control, vision sensing, projection display, audio feedback, wireless connectivity, and power management to ensure reliable standalone operation in desktop scenarios. The main components include: (1) Main Controller: Raspberry Pi Zero 2 W (512MB RAM) serves as the core computing and control unit, capable of running basic vision and interaction logic (with OpenCV support); (2) Micro Projector: a DLP2000-based module with 854×480 resolution and short-throw projection, used to project the interactive UI onto flat surfaces (desk/wall) to form an interaction area; (3) Camera: an OV5640 camera (5 MP, autofocus supported) captures gestures, objects, and environmental cues, enabling gesture recognition, interface registration, and task execution; (4) Speaker: a compact speaker module with PWM audio output provides sound cues and affective feedback; (5) Power System: two 18650 Li-ion cells (capacity ≥2000 mAh) target a battery life of at least 1 hour for demos and mobile usage; (6) Communication: Wi-Fi 2.4 GHz and Bluetooth 4.2 enable connection with a phone or external devices for control, debugging, and data transfer. 3. Software Modules The software modules integrate “projection display—visual perception—interaction comprehension—motion execution” into a closed-loop operational system. Employing a modular design for parallel development and future expansion, it comprises the following submodules: (1) Gesture Recognition Module: Implements gesture detection and recognition using MediaPipe/OpenCV, supporting inputs such as tap, thumbs-up, thumbs-down, and heart gestures for interaction commands. (2) Interface Rendering Module: Dynamically generates and renders main and sub-interface content (e.g., weather, clock), outputting corresponding graphical interfaces to the projection display. (3) Interaction Logic Engine: Maps gestures to commands and triggers events, manages interface state machines and interaction flows (main/sub-interface switching, exit/return), ensuring consistent and maintainable interaction logic. (4) Image Correction Module: Performs geometric correction and alignment on projected images to enhance stability at varying angles and distances. Integrates with cameras to implement auto-focus/alignment strategies, ensuring clearer and more reliable interface display. (5) Sound Effect Generation Module: Plays corresponding audio cues (e.g., thumbs-up, thumbs-down, heart feedback sounds) based on interaction events, providing clearer feedback. (6) Data Acquisition Module: Retrieves real-time weather, time, and other information via network APIs, updates projected interfaces, and enables information lookup functionality. (7) Motion Control Module: Manages chassis movement control and task execution, including fundamental speed/attitude control interfaces and higher-level behaviors like line-following navigation and moving to designated zones. This module integrates with the interaction logic engine, allowing users to trigger motion-related tasks via projected interfaces or gestures. #Criterion for success: ## F1: Main UI Projection Clarity & Icon Size • Success Criteria: The main interface projects two clearly visible icons (Weather and Clock). Each icon has a visible size of at least 3 cm × 3 cm. • Verification Method: Visually check projection clarity and measure the icon size using a ruler. ## F2: Weather Page Switching Latency • Success Criteria: After the user completes a click on the Weather icon, the UI switches to the weather page within 2 s. • Verification Method: Time the interval from click completion to page switch completion. ## F3: Weather Information Field Completeness • Success Criteria: The weather page displays, at minimum, the following fields: city, temperature, and weather condition. • Verification Method: Visually verify the presence of these fields on the projected page. ## F4: Clock Page Switching Latency • Success Criteria: After the user completes a click on the Clock icon, the UI switches to the time page within 2 s. • Verification Method: Time the interval from click completion to page switch completion. ## F5: Time Display Format & Refresh • Success Criteria: The time page displays the current time in “HH:MM:SS” format and updates continuously. • Verification Method: Visually check the format and observe continuous time updates. ## F6: Return/Exit Entry Consistency on Sub-Pages • Success Criteria: Sub-pages (Weather/Time) always show an Exit/Back icon (or an equivalent return entry) with a consistent, recognizable placement. • Verification Method: Visually check that the return entry remains present and consistent across pages. ## F7: Return-to-Main Page Latency • Success Criteria: After clicking the Exit/Back icon, the UI returns to the main page within 1.5 s. • Verification Method: Time the interval from click completion to main page display completion. ## F8: Thumbs-Up Gesture Response & Feedback • Success Criteria: Upon a thumbs-up gesture, the system displays a 👍 animation and plays a cheerful sound cue, with a total response time under 2 s. • Verification Method: Record a video and measure the latency frame-by-frame from gesture completion to animation/audio onset. ## F9: Thumbs-Down Gesture Response & Feedback • Success Criteria: Upon a thumbs-down gesture, the system displays a 👎 animation and plays a sad sound cue, with a total response time under 2 s. • Verification Method: Record a video and measure the latency frame-by-frame from gesture completion to animation/audio onset. ## F10: Heart Gesture Response & Feedback • Success Criteria: Upon a heart gesture, the system displays a ❤️ animation and plays a warm sound cue, with a total response time under 2 s. • Verification Method: Record a video and measure the latency frame-by-frame from gesture completion to animation/audio onset. ## F11: Gesture Recognition Accuracy • Success Criteria: Gesture recognition accuracy is at least 85% (at least 20 trials per gesture type). • Verification Method: Log the number of correct recognitions and total trials per gesture, then compute accuracy. ## F12: False-Trigger Rate (Robustness) • Success Criteria: False-trigger rate is no more than 10% (no response should be triggered by non-target gestures or no-interaction motions). • Verification Method: During a fixed-duration or fixed-count negative test (non-gesture/disturbance motions), record false triggers and compute the rate. |
||||||
| 29 | Interactive Projection System on Arbitrary Surfaces |
Jie Xu Jing Weng Yuqi Tang Zibo Dai |
design_document1.pdf design_document2.pdf final_paper1.pdf final_paper2.pdf other1.pdf |
Liangjing Yang | ||
| # Problem Most current smart devices rely on fixed-size screens for human-computer interaction, which limits display area, temporary collaboration, and natural input. Projection technology can extend interfaces into the physical environment, but conventional projectors usually provide visual output only and cannot support stable direct touch interaction across surfaces with different shapes, sizes, and materials. Our project aims to develop a system that projects an interactive user interface onto arbitrary physical surfaces and supports direct touch input on the projected area. This is a meaningful and technically challenging problem because the system must address not only projection, but also surface detection, projector-sensor calibration, touch localization, and real-time interaction feedback. We will begin by validating the first prototype on a normal wall, and then extend the design toward more general surfaces such as desks, paper, and other physical objects. # Solution Overview We propose to build an interactive projection system that integrates projection hardware, vision-based sensing, and embedded control. A projection module will display a graphical user interface on the target surface, while a camera or depth-based sensing module will monitor the surface and detect the position of a user’s finger during interaction. The sensed position will then be mapped into the projected interface coordinate system so that the system can recognize basic actions such as clicking and dragging, forming a complete display-sensing-recognition-feedback loop. The first implementation will be validated on a flat and stable wall surface; however, the overall architecture will be designed for extension to arbitrary surfaces, with attention to surface size variation, pose variation, and adaptive interface placement. Prior research shows that the key technical problems of arbitrary-surface interactive projection include surface segmentation and tracking, projector-camera calibration, and interaction area definition, which directly motivates our design. # Solution Components The proposed system consists of the following major components: ## Projection Display Module Projects a graphical user interface onto the target surface and adjusts the displayed area according to surface size, position, and orientation. ## Surface Sensing Module Uses a camera or depth/vision sensor to capture image or depth information from the target surface, detect surface geometry, and identify the available interactive area. ## Touch Detection and Interaction Recognition Module Detects whether the user’s finger is touching the projected surface and recognizes basic interaction events such as tapping and dragging. ## Coordinate Calibration and Mapping Module Establishes the spatial relationship between the sensing system and the projector so that detected touch points can be accurately mapped to interface locations. ## Embedded Control and System Integration Module Executes control logic, coordinates sensing and projection data flow, and manages communication and power across the system. ## Mechanical Support Structure Provides stable mounting for the projector, sensors, and control hardware so that the relative geometry remains fixed and repeatable during calibration and testing. # Criteria of Success The project will be considered successful based on the following criteria. 1. The system must project a stable and visible interactive interface onto at least one physical surface and maintain usable operation during demonstration. 2. It must detect direct touch input within the projected area and correctly trigger at least one basic interaction event, such as a click. 3. The touch localization accuracy must be sufficient for users to complete simple interface tasks such as button selection or menu navigation. 4. The system must demonstrate extensibility toward arbitrary surfaces by supporting interaction on at least one additional surface beyond a wall. 5. The complete prototype must support a demonstrable application scenario, such as a numeric keypad, simple control panel, or menu-based interface, showing that the full interaction loop has been implemented. These success criteria match the course expectation that requirements should be clear and verifiable, and they are also consistent with prior evaluation methods for click detection and drag interaction in projected interactive systems. |
||||||
| 30 | Automated Microwave Scatterometer and its digital twin |
Jianing Xiao Keyi Jin Yurong Wang |
final_paper1.pdf final_paper2.pdf final_paper3.pdf other1.pdf |
|||
| # Team Members: Yurong Wang (yurongw2) Keyi Jin (keyijin2) Jianing Xiao (xiao36) # Problem Overview Traditional microwave scatterometer systems rely heavily on manual or semi-automatic operation for calibration, antenna positioning, and measurement execution. This manual dependency introduces human error, limits experimental repeatability, and reduces overall efficiency. Current systems lack real-time synchronization between physical instruments and digital interfaces, preventing remote monitoring and intuitive control. Furthermore, physical hardware occupation during routine measurements blocks parallel activities such as algorithm development and experiment pre-planning. The absence of simulation capabilities restricts researchers from conducting "what-if" scenario analysis without consuming valuable instrument time. These limitations collectively hinder the advancement toward intelligent, autonomous electromagnetic measurement platforms required by modern aerospace and wireless communication applications. # Solution Overview Our solution for advancing traditional scatterometer systems is a Digital Twin. The Digital Twin for the microwave scatterometer can creates a seamless bridge between the physical and cyber worlds. It integrates hardware and software to enable automated data collection from the physical scatterometer instead of virtual simulation. The Digital Twin can automate calibration and measurement sequences, replacing manual operation with reliable, scheduled tasks. It can also create a real-time 3D visualization of the scatterometer's state within a model of its physical environment, allowing users to control the physical instrument through this interactive virtual model. # Solution Components ## Interactive visualization Subsystem - Real-time 3D scatterometer status display - Interactive interface for users to enter commands - Interface to display the data from the scatterometer ##Scatterometer Hardware Subsystem - Sun and Solar Panel for charging - Li-Ion Charger for emergency - Storage Battery to store the electricity provided by solar panels and chargers - Sensors used to detect the temperature and the humidity of the soil - Camera to capture data - Coupler and Wireless information transfer model to transder data between the device and the digital twin # Criterion for Success Automation & Control: The system shall execute fully automated calibration and measurement sequences with task scheduling reliability ≥99%, completely replacing manual intervention for routine operations. Measurement repeatability shall achieve coefficient of variation <2% across ten consecutive runs. Real-time Visualization: Unity-based 3D model synchronization latency shall not exceed 500 ms from physical sensor update to virtual state reflection. Antenna orientation accuracy shall be within ±1°, rotation speed within ±2% of setpoint. Simulation Accuracy: Virtual measurement results shall demonstrate trend consistency with physical measurements, enabling reliable "what-if" scenario planning for experiment design without hardware occupation. |
||||||
| 31 | Mobile eVTOL Handling and Docking Platform |
Carol Xu Haowen Chen Shu Yang Yuchen Zhang |
design_document1.pdf final_paper2.pdf |
Meng Zhang | ||
| #Problem Current quad-rotor eVTOL aircraft require a ground handling method that can safely support lifting, short-distance transfer, and accurate docking during parking, storage, and maintenance operations. Manual handling or improvised support equipment can lead to poor positioning accuracy, unstable lowering, inefficient turnaround, and increased risk of damage to the aircraft structure during ground operations. For repeated use in confined service environments, a dedicated ground support platform is needed to move the aircraft smoothly from its parking stand to a designated storage bay or maintenance station and place it in a stable, controlled, and repeatable manner. #Solution Overview The eVTOL Ground Support Lift-and-Transfer System is a ground support platform designed for quad-rotor eVTOL aircraft to enable lifting, short-distance transfer, and precise placement during ground operations. The system lifts the aircraft from its parking position using a motor-driven lifting mechanism and securely supports it during handling. A wheeled mobile platform allows the aircraft to be smoothly transported to a designated storage bay or maintenance station. Alignment guides and positioning stops assist with accurate docking, ensuring stable and repeatable placement. An Arduino-based control system coordinates lifting, movement, and safety monitoring to achieve controlled, reliable, and safe ground handling of the eVTOL aircraft. #Solution Components ##Lifting Subsystem -Load-bearing support structure for holding the eVTOL during handling -Motor-driven lifting mechanism for controlled raising and lowering -Contact interface that securely engages the aircraft landing structure -Travel limit detection to prevent overextension during vertical motion -Arduino-based control logic for stable lift and lower operation ##Mobility Subsystem -Wheel and drive assembly for smooth movement between locations -Braking or locking feature to keep the platform stationary during lifting and docking -Low-speed motion control for safe operation in confined ground environments ##Docking and Positioning Subsystem -Alignment guides for directing the eVTOL into the target position -Positioning stops for repeatable final placement -Sensors for detecting docking status and placement alignment -Feedback-based control sequence for precise and stable final positioning ##Central Control and Safety Subsystem -Arduino microcontroller for coordinating lifting, transfer, and docking actions -Sensor input processing for position, motion, and limit monitoring -Emergency stop and motion interlock logic for operational safety #Criteria of Success The system will be considered successful if it can safely lift a quad-rotor eVTOL aircraft from its parking position, transport it over a short ground distance, and accurately place it at a designated storage or maintenance location. The lifting mechanism must provide stable vertical motion without structural instability, while the mobility platform must enable smooth and controlled movement. The docking system should allow repeatable and precise positioning of the aircraft. In addition, the control system must reliably coordinate lifting, movement, and safety functions, including limit detection and emergency stop capability, ensuring safe and stable operation throughout the handling process. |
||||||
| 32 | Autonomous Target-Following Quadcopter with Real-Time YOLO Vision and Custom Flight Controller |
Jintu Guo Renang Chen Zhenbo Chen Zhengyu Zhu |
design_document1.pdf design_document2.pdf final_paper2.pdf other1.pdf other2.pdf proposal1.pdf |
Lin Qiu | ||
| # Problem Small unmanned aerial vehicles (UAVs) are widely used in applications such as aerial filming, search-and-rescue, and surveillance. However, most consumer-grade FPV (First Person View) drones rely entirely on manual control and lack the ability to autonomously track moving targets. We aim to design an autonomous target-following quadcopter system that leverages edge computing for real-time object detection. The drone needs to recognize a specific target using a vision system and autonomously follow it while maintaining stable flight, bridging the gap between manual FPV drones and expensive enterprise autonomous platforms. # Solution Overview Our solution consists of a custom-built 5-inch quadcopter (using a Mark frame, 2207 brushless motors, and 55A ESCs) equipped with an Orange Pi 5 as the central vision processor, and a custom-designed Flight Controller PCB. The Vision Subsystem on the Orange Pi 5 will run a YOLO object detection model to capture and identify the target's relative position. This spatial data is sent to our custom Flight Control Subsystem (STM32-based), which executes a closed-loop PID control algorithm to adjust the drone's attitude and thrust. To meet the high power demands of the vision board, the Power Subsystem—integrated into our custom PCB—will feature an optimized three-level buck converter to safely and efficiently step down the high-voltage LiPo battery to a stable 5V/4A supply. A Remote Control Subsystem will allow FPV manual override and mode switching. # Solution Components ## Power Subsystem A custom PCB integrating an advanced three-level buck converter. It steps down the voltage from a high-capacity LiPo battery (sized appropriately to target a 15-minute flight time) to provide a stable, low-ripple 5V/4A power supply for the Orange Pi 5, while routing raw power to the 55A Electronic Speed Controllers (ESCs). ## Flight Control Subsystem The core of our custom PCB, built around an STM32 microcontroller and an IMU (e.g., MPU6000/BMI270). It receives tracking vectors from the Vision Subsystem and user inputs from the receiver, generating precise PWM signals for the ESCs to stabilize the drone and follow the target. ## Vision Subsystem An Orange Pi 5 paired with a high-framerate camera module. It runs a YOLO-based object detection algorithm to process video feeds in real-time, computing the bounding box and relative spatial coordinates of the target object. ## Remote Control & Interaction Subsystem A wireless FPV radio receiver link that allows the operator to manually control the drone, monitor telemetry, and safely toggle between manual FPV flight mode and autonomous tracking mode. # Criterion for Success - The custom three-level buck converter on the PCB can stably output 5V at 4A under continuous load, and sustain a peak current of 5A for up to 10 minutes without requiring additional active cooling or resetting the Orange Pi 5. - The Vision Subsystem (Orange Pi 5) successfully runs the YOLO model at a minimum of 30 FPS to detect and output the relative coordinates of a target. - The flight controller can smoothly process vision data to autonomously follow a target moving at a walking pace (1-2 m/s), keeping the target within the camera's field of view for at least 15 seconds. - The customized power distribution and selected LiPo battery support a continuous flight/hover time approaching 15 minutes. - The remote control system allows seamless switching between autonomous tracking and manual FPV override with a control latency of less than 300 ms. # Distribution of Work - **Zhenbo Chen (EE):** Power Subsystem design. Responsible for the custom PCB schematic and layout of the high-efficiency three-level buck converter and power distribution to the ESCs. - **Zhengyu Zhu (EE):** Flight Control Subsystem design. Responsible for the STM32 integration on the custom PCB, IMU sensor fusion, and embedded PID flight control firmware. - **Renang Chen (ECE):** Hardware Assembly and Systems Integration. Responsible for the Mark frame mechanical build, 2207 motor/55A ESC integration, battery sizing/testing, and Remote Control Subsystem configuration. - **Jintu Guo (ECE):** Vision Subsystem design and implementation. Responsible for configuring the Orange Pi 5, deploying the YOLO model, and writing the serial communication protocol to send coordinate vectors to the STM32. # Justification of Complexity We believe our project possesses the significant electrical and embedded systems complexity required for ECE 445. The hardware core of this project is a highly complex custom PCB that must integrate a sensitive STM32 flight controller alongside a high-current, high-efficiency three-level buck converter. Delivering a clean 5V/4A to the Orange Pi 5 in an extremely noisy FPV drone environment (caused by 55A ESCs and high-KV 2207 motors) requires rigorous PCB layout, impedance matching, and thermal management skills. Furthermore, developing the embedded C firmware on the STM32 to bridge FPV radio inputs with autonomous YOLO-derived spatial vectors from a Linux board involves advanced knowledge of control theory, sensor fusion, and real-time communication protocols. |
||||||
| 33 | Automated Homemade Dog Food Production Machine |
Jingyang Chen Wenkai Zheng Zekai Song Zixi Zhao |
design_document1.pdf final_paper1.pdf final_paper2.pdf other1.pdf other2.pdf other3.pdf other4.pdf video1.mp4 |
Fangwei Shao | ||
| #Problem Many dog owners prefer to prepare homemade food for their pets using fresh ingredients such as vegetables and meat. However, manually preparing dog food is time-consuming and inconsistent. Ingredients must be chopped, mixed, shaped, and dried to produce stable dog food pieces suitable for storage and feeding. Existing kitchen appliances such as blenders or food processors can perform only part of this process, requiring users to manually transfer and process ingredients multiple times. There is currently no compact system designed specifically to automate the entire pipeline of preparing and forming homemade dog food pellets from raw household ingredients. Therefore, a system that can automatically process common ingredients and convert them into dry dog food pellets would significantly simplify homemade pet food preparation. #Solution Overview Our solution is an automated machine that converts raw household ingredients into dry dog food pellets through a sequence of mechanical processing stages. First, raw ingredients such as vegetables, meat pieces, or ground meat are placed into a feeding chamber. A rotating blade mechanism chops and grinds the ingredients into smaller particles. The processed mixture then enters a cylindrical mixing chamber containing a screw auger that continuously mixes and transports the material. The mixture is then extruded through an outlet where a rotating cutting blade divides the extruded material into small pellet-like pieces. These pellets fall onto a tray where a hot-air drying system removes moisture and solidifies the food into stable dog food pieces. A microcontroller-based control system will coordinate the motors and heating elements to regulate the chopping, mixing, extrusion, cutting, and drying processes. #Solution Components [Ingredient Processing Subsystem] Chopping blade module: A rotating blade driven by an electric motor to chop raw ingredients into smaller particles. Mixing and transport module: A cylindrical chamber with a screw auger that mixes ingredients and transports them toward the extrusion outlet. [Pellet Formation Subsystem] Extrusion outlet: A nozzle structure that shapes the mixed food material into a continuous output stream. Rotating cutting blade: A motor-driven blade that cuts the extruded material into small pellet-sized pieces. [Drying Subsystem] Hot-air drying module: A heating element and fan system that circulate warm air to remove moisture from the pellets and form stable dog food pieces. [Control Subsystem] Microcontroller unit: Controls motor operation, timing of each processing stage, and temperature of the drying system. #Criterion for Success The chopping and mixing mechanism must produce a sufficiently uniform mixture suitable for extrusion. The extrusion and cutting system must produce pellets of relatively consistent size. The drying subsystem must remove sufficient moisture so that the pellets maintain their shape and can be stored without immediate spoilage. The system must operate automatically once ingredients are loaded, coordinating the chopping, mixing, extrusion, cutting, and drying processes. |
||||||
| 34 | A Vision-Integrated Robot for Autonomous Book Classification in Library Environments |
Xinrui Xiong Zehao Bao Zhecheng Lou Zhenxiong Tang |
design_document1.pdf final_paper1.docx final_paper2.pdf other1.pdf other2.pdf other3.pdf other4.pdf video1.mp4 |
Timothy Lee | ||
| Problem In library operations, staff must identify returned books, look up shelf locations, and reshelve by hand—a labour-intensive process that does not scale and is harder when collections include multiple languages. We focus on a fixed workstation (no mobile base): a robotic arm picks books from a collection bin and places them on the shelf. The real challenge is not “place at (x,y,z)”, but how the robot visually understands the current bookshelf state and decides where and how to place each book—perceiving gaps, widths, and fit, then choosing stable, tidy placements. This “shelf perception and placement decision” remains active in warehouse robotics. We concentrate innovation on this software side and simplify hardware to a fixed arm, aiming for clear, publishable contributions aligned with “software tuning”. Solution Overview Fixed workstation: collection bin, arm with gripper, and shelves within reach. Readers place books in a vertical bin with fixed spine (barcode) orientation. The system runs in batches (scan barcodes, look up positions, plan return order by greedy or similar). Innovation in three areas: (1) Real-time shelf occupancy perception—camera scans each layer, image analysis finds gaps and widths and whether the current book fits. (2) Intelligent slot selection—policy-based choice (prefer larger gaps, avoid squeezing and isolated positions). (3) Visual-servo placement—camera gives real-time feedback during insertion instead of open-loop trajectories. One-button operation; runs on a single-board computer such as Raspberry Pi. Demonstrator uses a scaled-down environment and lightweight mock books. Solution Components Perception Subsystem (Subsystem I) -Barcode/QR scanner or camera for book identity; optional depth camera for pose. Barcode decoding library and catalogue interface for shelf (level, slot). -Shelf occupancy (innovation): Camera scans a layer; image analysis returns gap locations, widths, and fit for the current book. Output to slot selection. -Output: book identity and target range; per-layer gap/occupancy for placement. Task Planning and Scheduling Subsystem (Subsystem II) -Batch order by simple greedy (stack order or shelf proximity). Intelligent slot selection (innovation): Within correct range, select slot by policy—larger gaps, next to similar books, avoid isolated positions. Designed for algorithm design and quantitative evaluation. -Output: ordered task list (book → shelf, level, slot) from shelf perception and policy. Manipulation Subsystem (Subsystem III) -Arm (open-source 6-DOF) with smart servos (serial bus), gripper, and on-board camera. Visual-servo placement (innovation): Real-time camera feedback adjusts pose/trajectory during insertion (closed-loop “software tuning”). Kinematics and trajectory planning; camera confirms stability after placement. -Output: book in chosen slot, stable (upright or leaning), vision-verified. Criterion for Success 1. One-button operation: load bin, press start; system runs scan, planning, placement, then ready for next batch. 2. Barcode accuracy ≥ 95%; full flow from scan to reshelve without human intervention. 3. Success = book in correct range, upright or leaning; placement success rate ≥ 90%. 4. The three innovations in use: visual shelf occupancy for placement; intelligent slot policy (observable or evaluable); visual feedback during placement (not only open-loop). 5. No serious hardware collisions; shelf left tidier where feasible. Demo success: Load a batch (five books), press once; system completes identification, planning, and placement using shelf perception, intelligent slot choice, and visual-servo placement; every book in correct range and stable, no serious collisions. |
||||||
| 35 | Autonomous Ammunition Loading and Firing Robotic System |
Xiaoman Li Xinchen Yao Yidong Zhu Yuxuan Nai |
design_document1.pdf final_paper1.pdf final_paper2.pdf photo1.jpg proposal1.pdf |
|||
| # Problem In competitive robotics (such as RoboMaster) and automated defense systems, rapidly and accurately reloading ammunition in dynamic environments is critical. Traditional hard-coded reloading mechanisms lack adaptability; they often fail or jam when the ammunition (e.g., darts) is not perfectly aligned in the staging area or when the system faces mechanical disturbances. Currently, there is a need for a highly adaptable, intelligent manipulation system that can autonomously locate, grasp, and load projectiles from unstructured staging zones into a firing mechanism without human intervention. # Solution Overview Our solution is an autonomous manipulation and launching system featuring a RoboMaster-inspired robotic arm integrated with a dart launcher. The system is initiated via a user-friendly trigger (physical button or remote command). Once activated, the system uses an onboard vision setup to observe the staging area behind it. To achieve robust manipulation, we will deploy an adaptable, intelligent decision-making policy. Depending on evaluation results during development, this overarching policy may take the form of an end-to-end Vision-Language-Action (VLA) model, a reinforcement learning (RL) agent trained via simulation-to-reality transfer, or a modular vision-based state estimation pipeline paired with adaptive planning. Guided by this policy, the arm will autonomously calculate the optimal trajectory to grasp the dart, navigate to the loading port, secure the dart into the mechanism, and execute the launch sequence. # Solution Components * **Perception and Decision Subsystem:** Consists of the onboard camera(s) and the central compute unit running the overarching intelligent policy (e.g., vision-based agent, VLA model, or detection script) to perceive the dart's location and generate the corresponding action sequence. * **Control and Planning Subsystem:** Translates the high-level policy outputs into low-level motor commands, calculating inverse kinematics and generating smooth, collision-free trajectories for the arm to move from the rear staging area to the front loading mechanism. * **Actuation and Launch Subsystem:** The mechanical hardware, including the RoboMaster-inspired robotic arm (servos and gripper) for handling the dart, and the integrated firing mechanism/actuator responsible for launching the projectile. * **Power and Interface Subsystem:** Includes the power management circuits that supply stable voltage to the motors, launcher, and microcontroller, along with the physical/remote initiation interfaces. # Criteria of Success * The system must successfully initialize and begin the observation and retrieval sequence reliably when triggered. * The perception/policy module must successfully identify the presence and general location of the dart in the staging area behind the robot. * The robotic arm must successfully grasp the dart, navigate to the loading port, and secure it in the launcher without dropping it. * The launcher mechanism must successfully activate and fire the loaded dart towards a general forward direction. |
||||||
| 36 | Intelligent Basketball Retrieval & Return Robot |
Jinghui Zheng Libo Zhang Linzhi Du Zichao Lin |
design_document1.pdf final_paper1.pdf other1.pdf other2.pdf other3.pdf presentation1.pptx presentation2.pptx |
Timothy Lee | ||
| 1. Problem 1.1 Background Basketball training often requires players to repeatedly retrieve the ball after each shot, which interrupts practice rhythm and reduces training efficiency. Although some existing training machines can return the ball automatically, many of them only deliver passes in fixed directions and cannot adapt to the player’s changing position. Therefore, there is a practical need for a smarter basketball passing device that can collect the ball automatically, locate the user, and return the ball accurately to support more efficient and continuous individual training. 1.2 Problem Statement The problem this project aims to address is how to design and develop an automated basketball passing machine that can collect the ball after a shot, identify the user’s position through a wearable Bluetooth locator, and deliver the ball back to the user with appropriate direction and accuracy. The system should improve the continuity and efficiency of individual basketball training. 2. Solution overview The proposed solution is an automated basketball passing machine that integrates mechanical design, electronic hardware, and embedded control to achieve ball collection, user localization, and directional ball delivery. After the basketball enters the machine, a collection and feeding mechanism transfers it into a ready-to-launch position. At the same time, the system receives location information from a wearable Bluetooth device attached to the user’s wrist. Based on the processed positioning data, the controller estimates the user’s relative direction and distance, and then adjusts the launching mechanism accordingly. From a technical perspective, the system consists of three main parts. First, the mechanical subsystem includes the ball collection structure, ball storage and feeding mechanism, and a launching unit capable of controlling the release direction and passing speed. Second, the electronic subsystem includes the main control board, Bluetooth communication module, motor drivers, sensors, and power management circuit, which together support signal acquisition and actuator control. Third, the software subsystem is responsible for localization data processing, motion control, and coordination of the overall operating sequence. Through the integration of these subsystems, the machine is expected to provide a more intelligent and efficient solution for continuous individual basketball training. 3. Components 3.1 Mechanical Structure The mechanical structure of the proposed basketball passing machine can be divided into three main parts: the ball collection mechanism, the ball storage and feeding mechanism, and the ball launching mechanism. First, the ball collection mechanism is designed to capture the basketball after the user takes a shot and guides it back into the machine automatically. A net-based collection structure, similar to a funnel-shaped mesh, is installed around the basket and the main rebound area. This net is supported by a lightweight frame and connected to the inlet of the passing machine. After the ball falls or rebounds from the basket area, it is intercepted by the inclined net surface and rolls downward under gravity toward the collection opening. In this way, the system can reduce manual ball retrieval and improve the continuity of shooting practice. Second, the ball storage and feeding mechanism adopts an inclined rolling channel. After entering the machine inlet, the basketball moves into a sloped channel where multiple balls can be temporarily stored in sequence. Because of the inclined geometry, each ball rolls naturally toward the feeding end under gravity. A controllable blocking device is installed near the launching position to ensure that only one ball is released at a time. This design is mechanically simple, easy to manufacture, and suitable for stable sequential ball feeding. Third, the ball launching mechanism uses a dual-wheel launching structure. In this design, two high-speed rotating wheels are arranged on both sides of the basketball. When a ball is fed into the launching position, the friction generated by the rotating wheels accelerates the ball and launches it toward the user. By adjusting the rotational speed of the wheels, the system can control the passing speed and adapt to different user distances. Compared with other launching methods, the dual-wheel structure offers better controllability, relatively simple construction, and good compatibility with motor-driven actuation. To further improve the directional passing capability, two additional adjustment mechanisms are included in the launching subsystem. The first is a launching angle adjustment mechanism, which allows the launcher to change its vertical angle so that the ball trajectory can be adapted for different passing distances or heights. The second is a turntable base, which supports horizontal rotation of the entire launching unit. Driven by a motor, this rotating base enables the machine to align the launcher with the user’s position based on the localization result. With the combination of vertical angle adjustment and horizontal rotation, the machine can achieve more flexible and accurate ball delivery. 3.2 Electronic Hardware The electronic hardware system is responsible for sensing, computation, communication, and actuation control of the robot. It mainly consists of the following modules: Main Control Board: The main control board (microcontroller-based) serves as the central processing unit of the system. It coordinates all subsystems, processes sensor data, calculates motion and launching parameters, and sends control signals to the motor drivers and actuators. Wireless Positioning Module: A wireless positioning module (such as Bluetooth beacon or UWB module) is used to estimate the real-time location of the trainer. The robot receives positioning signals from the wearable device and determines the relative position of the user, enabling accurate ball return direction calculation. Motor Driver Modules: Motor driver circuits are used to control the motors responsible for ball collection, internal ball transport, and the launching mechanism. These drivers provide sufficient current and precise speed control to ensure stable mechanical operation. Power Management Module: The power module regulates and distributes power from the battery to different electronic components. Voltage regulation circuits ensure stable operating voltages for the microcontroller, sensors, communication modules, and motor drivers. Sensors and Interface Circuits: Additional sensors and interface circuits may be used to detect system states such as ball presence, motor status, or mechanical limits. These sensors help the controller monitor system operation and improve reliability and safety. 3.3 Software The software subsystem serves as the central intelligence of the basketball passing machine, responsible for localization data processing, motion control, and coordination of the overall operating sequence. First, the positioning information processing module interprets data received from the user's wearable Bluetooth locator to accurately determine the player's real-time position, distance, and relative angle. Second, based on these coordinates, the launch parameter calculation algorithm determines the necessary mechanical adjustments. This includes calculating the required rotational speed for the dual-wheel launching structure to achieve the correct passing speed , as well as computing the target angles for the horizontal turntable and vertical adjustment mechanisms. Third, the motion control program translates these computed parameters into precise electrical signals to drive the respective motors accurately. Finally, the overall system control logic acts as the main state machine that seamlessly manages the entire workflow—from coordinating the controllable blocking device for ball feeding to target locking and launching—ensuring safe, continuous, and reliable operation. 4. Criteria of Success 4.1 Functional Success The system must be able to autonomously collect basketballs that land within the operating area near the hoop and return them to the trainer without manual intervention. After detecting the presence of a basketball, the robot should navigate to the ball, retrieve it using the ball collection mechanism, and transport it to the launching module. Using the wireless positioning information from the trainer’s wearable device, the system must determine the relative position of the user and adjust the launching mechanism accordingly. The robot should then launch the basketball toward the trainer so that it can be received within a reasonable catching distance. Throughout the process, the system should perform ball detection, retrieval, positioning, and launching in a coordinated and automated manner, enabling continuous basketball training with minimal interruption. 4.2 Performance Success The system must achieve stable and reliable performance during autonomous basketball retrieval and return operations. The robot should be able to successfully collect basketballs located within the designated operating area near the hoop and transport them to the launching mechanism without jamming or mechanical failure. When returning the ball to the trainer, the launching mechanism should deliver the basketball within a reasonable catching distance of the user. The target landing area should be within approximately 1–2 meters of the trainer's position, allowing the trainer to receive the ball comfortably during practice. Additionally, the system should complete the entire cycle of ball detection, retrieval, positioning, and launching within a reasonable time, enabling continuous training without long interruptions between shots. The robot should also maintain consistent operation over multiple retrieval cycles to demonstrate system stability and reliability. 4.3 Engineering Success Engineering success for this project is defined by the complete design, fabrication, and integration of its three primary technical disciplines: mechanical structure, printed circuit board (PCB) hardware, and software code. From a mechanical standpoint, this requires the physical realization and stable assembly of the collection, feeding, and launching mechanisms. For the electronic hardware, it necessitates the development of a fully functional custom PCB that reliably integrates the main control board, Bluetooth communication module, motor drivers, sensors, and power management circuits. For the software, it requires the deployment of robust control algorithms and communication protocols. The ultimate criterion for engineering success is the successful joint debugging (system integration) and continuous operation of the complete machine. This final milestone will demonstrate that all mechanical components, electronic circuits, and embedded code work together seamlessly as a unified system to achieve automated ball collection, user localization, and directional delivery. |
||||||
| 37 | Intelligent Waste Sorting System |
Canyu Li Han Yin Mingyang Gao Wentao Li |
design_document1.docx final_paper1.pdf other1.pdf other2.pdf |
Bo Zhao | ||
| Intelligent Waste Sorting System ##Team -Wentao Li(wentaol5) -Canyu Li(canyuli2) -Hanyin(hanyin3) -Mingyang Gao(mg82) ##Problem With rapid urbanization, the volume of domestic waste has surged, making correct waste sorting a crucial part of environmental protection and resource recycling. However, current waste sorting in public areas relies heavily on public awareness, leading to a high error rate in source sorting. Incorrect sorting increases the cost of manual processing and can cause entire bins of recyclables to be contaminated and sent to landfills. While large treatment plants have automated sorting lines, there is a lack of compact, efficient, and low-cost automated sorting equipment at the source of waste generation (e.g., schools, office buildings). We need a smarter way to lower the barrier for ordinary people to sort waste, using machines instead of manual labor for initial screening to improve resource recovery rates. ##Solution Overview Our solution is to design and build an intelligent waste bin integrating computer vision and a fixed mechanical device, deployed in public areas. Users simply throw their waste into a unified drop-in opening. A camera inside the system captures images of the items and uses a pre-trained machine learning model to identify and classify them (e.g., Recyclables, Food Waste, Hazardous Waste, Other Waste). Once identified, a microcomputer triggers the fixed mechanical device (such as servo-driven baffles or push rods) to automatically guide and sort the item into the corresponding internal waste bin. This provides a fully automated, touchless waste disposal process, greatly improving the accuracy and efficiency of source sorting. Solution Components ##Vision and Control Subsystem Image Capturing Module: A high-definition camera deployed in the waste drop-in channel to capture clear images while the waste is temporarily stationary or falling slowly. Processing & Control Unit: A microcomputer (e.g., Raspberry Pi or Jetson Nano) to receive images, run the computer vision model (object detection |
||||||
| 38 | Dual-Arm Robotic System for Cube Rotation |
Keeron Huang Rong Wang Yiming Xu Zhuoyang Shen |
design_document1.pdf final_paper1.pdf final_paper2.pdf proposal1.pdf |
Meng Zhang | ||
| # Dual-Arm Robotic System for Cube Rotation Team members (listed A-Z): - Qixuan Huang - Rong Wang - Yiming Xu - Zhuoyang Shen ## Problem Traditional Rubik’ cube solvers often rely on highly specialized, single-purpose mechanical structures that lack the versatility of human-like manipulation. Conversely, general-purpose bimanual robots struggle with the precision required for cube rotation and the complex coordination needed to prevent jamming. Furthermore, training robust bimanual policies requires massive amounts of data; collecting this in the real world is time-consuming, expensive, and risks damaging expensive hardware. There is a need for a system that leverages advanced simulation data to perform high-precision, robust bimanual manipulation in the physical world. ## Solution Overview We propose an integrated bimanual robotic system that uses RoboTwin 2.0 to bridge the gap between simulation and reality (Sim-to-Real). - Simulation & AI: We will utilize RoboTwin 2.0’s “Strong Domain Randomization” (varying lighting, clutter, and textures) to generate a massive synthetic dataset. This data will be used to train a robust bimanual manipulation policy capable of handling physical uncertainties. - Hardware Implementation: To meet the course's hardware requirements, we will construct a physical dual-arm workstation. The system will feature a custom-designed PCB for power distribution and motor control, ensuring the mechanical arms can execute the trained policy with high torque and precision. - User Interface: The system will adhere to the "One-button start" requirement, where a single physical trigger initiates the vision-scan-solve-rotate sequence autonomously. ## Solution Components - Subsystem I: Mechanical & Actuation (ME Focus) - Bimanual Arm Assembly: Two 3-to-6 DOF robotic arms equipped with specialized grippers. - 3D Printed End-Effectors: Custom-designed high-friction fingertips and cube-stabilizing fixtures to ensure secure grasping during high-speed rotations. - Subsystem II: Electronics & Control (ECE Focus - Core Requirement) - Custom PCB: A dedicated circuit board integrating a voltage regulation module (12V to 5V/3.3V), high-current motor driver ICs (e.g., PCA9685 for PWM expansion), and signal isolation to protect the MCU. - Central MCU: An ESP32 or STM32 microcontroller to handle real-time motor commands and “One-button” logic. - Subsystem III: Vision & Computation - Sensing: A dual-camera or mirror-based vision system for 6-face color recognition. - Edge Computing: A Jetson Nano or PC to run the RoboTwin-trained policy and the Kociemba solving algorithm. ## Criteria of Success - Vision Accuracy: Correct identify the color configuration of all 6 faces of a scrambled cube within 30 seconds under varying ambient light. - Mechanical Stability: The bimanual arms must successfully rotate the cube faces without dropping the cube or causing mechanical jamming in 95% of test trials. - Full Autonomy: Upon pressing the physical start button, the system must autonomously solve the cube from any scrambled state within 3 minutes. - Hardware Integrity: The custom PCB must operate without overheating or voltage drops exceeding 5% during peak motor activity. ## Distribution of Work - Yiming Xu: Develops the simulation environment using RoboTwin 2.0 and is responsible for generating synthetic expert datasets and training the bimanual manipulation policy via domain randomization. - Zhuoyang Shen: Designs the computer vision module for Rubik’s cube state recognition and implements the high-level solving algorithm (e.g., Kociemba) for optimal motion path planning. - Rong Wang: Responsible for the custom PCB design and hardware implementation, including high-current motor drive circuits, power management systems, and low-level MCU firmware for real-time control. - Qixuan Huang: Focuses on the mechanical structure design and fabrication, including 3D-printed specialized bimanual grippers and performing system-wide sim-to-real integration and stability testing. |
||||||
| 39 | A Morphable Bionic Robotic Fish with Dual-Mode Propulsion Enabled by a Transformable Caudal Mechanism |
Bowen Zhang Kaijun Zheng Libin Wang Xuanyu Ke |
design_document1.pdf final_paper1.pdf final_paper2.docx video1.mp4 |
Hua Chen | ||
| # Problem Most existing underwater robotic fish rely on only one propulsion method, which limits their adaptability across different tasks and environments. Bioinspired tail propulsion is typically smooth, quiet, and energy-efficient, making it well suited for steady swimming and maneuvering. In contrast, propeller-based propulsion can provide higher speed, faster acceleration, and stronger turning capability. However, integrating both propulsion methods into a compact robotic fish remains technically challenging, since the tail structure, actuation mechanisms, and control strategies often impose conflicting design requirements. Therefore, there is a need for a transformable underwater robotic platform that can switch between different propulsion modes and better adapt to diverse operational scenarios. # Solution Overview Our project aims to develop a transformable underwater robotic fish capable of multi-mode propulsion. The system integrates a waterproof electronic enclosure, a bioinspired tail-swing actuation mechanism, and a caudal-fin-to-propeller morphing structure within a compact robotic body. In one mode, a servo-driven tail produces periodic left-right oscillations to achieve fish-like swimming behavior with smooth and efficient motion. In the other mode, the rear-end transformation mechanism re-configures the caudal fin into a propeller structure driven by a brushless motor to generate higher thrust for rapid movement. A manual control system is used to command both locomotion and propulsion mode switching. Through this design, the robotic fish provides a promising platform for studying adaptive underwater locomotion in different environments. # Solution Components ## Waterproof Electronic Enclosure Subsystem - A waterproof shell used to protect internal components from water ingress during underwater operation. - An embedded control PCB used to coordinate sensing, actuation, and system-level control. - A battery module used to provide onboard power for the robotic fish. - An IMU used to measure motion state and orientation during swimming. - Electrical and communication interfaces used for subsystem integration and testing. ## Bioinspired Robotic Fish Tail-Swing Actuation Subsystem - A set of 3D-printed mechanical structural components designed to support the tail-swing motion. - Servos used to drive the oscillatory motion of the robotic tail for bionic propulsion. ## Caudal Fin-to-Propeller Morphing Subsystem 3D-printed fin-like components designed to support both fin and propeller configurations. - A morphing mechanism used to transform the caudal fin into a propeller-based propulsion structure. - A brushless motor used to drive rotational thrust in propeller mode. ## Manual Control System - A controller used to manually command the robotic fish’s movement and switch between propulsion modes. # Criteria of Success The project will be considered successful based on the following criteria. ## Stable Underwater Locomotion The robotic fish must be able to perform basic underwater locomotion, including stable forward swimming and directional steering in a water environment. ## Reliable Mode Transformation The tail transformation mechanism must successfully switch between the bionic flapping mode and the propeller propulsion mode while submerged, and the structure must remain stable and mechanically reliable throughout the transition. Also, both propulsion modes can operate normally. ## Data Acquisition and Evaluation The system must be capable of collecting key operational data so that the performance of the two propulsion modes can be evaluated and compared. # Members and Work Distribution - Zhang, Bowen (ME): Mechanical design and assembly - Ke, Xuanyu (ME): Mechanical design and assembly - Zheng, Kaijun (EE): Electronic system design and testing - Wang, Libin (ECE): Control algorithm development and system programming |
||||||
| 40 | Offline Multi-Factor Authentication Smart Safe |
Ruichao Chen Ziheng Yu Ziyuan Luo |
design_document1.pdf final_paper1.pdf final_paper2.pdf final_paper3.pdf proposal1.pdf |
|||
| Project Name: Offline Multi-Factor Authentication Smart Safe Overview Traditional single-point authentication safes are vulnerable to key theft or password cracking. Our project is a standalone, battery-powered smart safe equipped with a three-in-one multi-factor authentication (MFA) system: edge AI facial recognition, fingerprint recognition, and RFID. This system is designed for secure and intuitive interaction, verifying biometric data locally and triggering the physical electromechanical lock instantly, without relying on vulnerable cloud networks or smartphone apps. Unique Features Unlike commercial smart locks that process data via Wi-Fi (which poses significant privacy and cybersecurity risks), our system performs all biometric matching entirely at the edge. Crucially, the system architecture is controlled by a strictly non-blocking finite state machine (FSM). This FSM logic supports a “high-security mode” (enforcing strict sequential multi-factor authentication), resisting brute-force attacks. These features are typically found in enterprise-grade security systems, not consumer-grade desktop safes. Brief Technical Overview The core component is a custom-designed PCB employing a dual-MCU architecture: an ESP32-S3 (or other hardware) handles the DVP camera interface and edge AI algorithms, while an STM32 manages the FSM and peripheral polling. The main challenge in the hardware design lay in building a robust power distribution network and high-current MOSFET drive circuitry. This ensured that the 12V electromechanical electromagnetic lock could safely withstand peak transient currents without causing voltage drops to sensitive logic circuitry. |
||||||
| 41 | OmniGrasp: VLA-Driven Mobile Manipulator with Custom-Built 7-DOF Arm and Mecanum Chassis |
Dayu Xia Shurong Wang Tongning Zhang Yaofang Ji |
design_document1.pdf final_paper1.pdf final_paper2.pdf other1.pdf |
Piao Chen | ||
| # Project Proposal for ECE 445 **Project Title:** OmniGrasp: VLA-Driven Mobile Manipulator with Custom-Built 7-DOF Arm and Mecanum Chassis **Team Members:** Tongning Zhang, Shurong Wang, Dayu Xia, Yaofang Ji ### Problem In complex daily or industrial environments, standard robotic manipulators are fundamentally limited by a stationary workspace and rigid, pre-programmed trajectories. Furthermore, commercial 7-DOF robotic arms are often prohibitively expensive, overly heavy, or difficult to seamlessly integrate with custom mobile platforms. There is a critical need for an intelligent, highly integrated mobile system that combines a custom-manufactured, lightweight manipulator with autonomous navigation and high-level reasoning capabilities driven by natural language commands. ### Solution Overview Our solution is a mobile robotic manipulator capable of navigating dynamic environments to execute complex, multi-stage grasping tasks. The hardware consists of an omnidirectional Mecanum wheel chassis paired with a custom-designed and manufactured 7-Degree-of-Freedom (7-DOF) robotic arm and a versatile gripper. To achieve optimal weight distribution and payload capacity, the 7-DOF arm will be fully modeled in CAD and fabricated in-house using a combination of 3D printing and precision machining. The system operates on a hierarchical control architecture. At the highest level, a Vision-Language-Action (VLA) model running on an edge computer interprets natural language commands and plans the necessary mobile navigation and manipulation sequences. The system seamlessly coordinates the base movement and the custom arm's inverse kinematics to approach and grasp target objects. ### Solution Components * **Mechanical Design and Fabrication Subsystem:** The physical structure of the 7-DOF arm, including custom-engineered links, joint enclosures, and motor mounts, designed via CAD software and manufactured using 3D printing and CNC machining to ensure structural integrity and optimal payload-to-weight ratio. * **Perception and High-Level Planning Subsystem:** An edge computer hosting a Vision-Language-Action (VLA) model designed to interpret natural language commands (e.g., "Navigate to the table and grasp the red cube") and plan comprehensive task sequences. * **Low-Level Control Subsystem:** A microcontroller dedicated to executing omnidirectional kinematics for the Mecanum chassis and inverse kinematics with PID motor control for the 7-DOF arm's precise movement. * **Mobility and Actuation Subsystem:** The integrated hardware platform comprising the Mecanum wheel base for unconstrained floor movement and the custom 7-axis robotic joint motors equipped with a gripper for dexterous physical manipulation. * **Feedback Subsystem:** A sensor feedback loop integrating vision or odometry for base positioning, alongside torque or current sensors designed to provide grasp verification during object manipulation. ### Criteria of Success * The custom-manufactured 7-DOF arm must maintain structural integrity under its designated maximum payload without significant mechanical deflection or structural failure. * The Mecanum chassis must successfully navigate to designated locations based on the VLA model's planning. * The high-level VLA model must accurately interpret natural language commands to plan correct, multi-stage task sequences involving both mobility and manipulation. * The 7-DOF robotic arm must demonstrate the ability to calculate and execute smooth trajectories to reach target objects, effectively utilizing its redundant joints. * The gripper must successfully grasp and manipulate various target objects based on their geometry without dropping them. |
||||||
| 42 | Low-Latency Analog Differential Equation Solver |
Jiachang Wang Tianyue Jia Yanzi Li Yishan Sheng |
design_document1.pdf final_paper1.pdf other1.pdf |
Aili Wang | ||
| # Problem Ordinary Differential Equations (ODEs) are widely used to describe dynamic systems in the real world, such as mechanical vibration systems, electrical circuits, and thermal processes. These systems evolve continuously over time, and their dynamic behaviors are commonly modeled using differential equations. For example, a typical mass–spring–damper system can be described as $mx+cx+kx=F(t)$ where $x(t)$ represents the displacement of the system, $m$ is the mass, $c$ is the damping coefficient, $k$ is the spring stiffness, and $F(t)$ represents the external force applied to the system. Traditionally, such differential equations are solved numerically using digital computers. Numerical methods such as Euler’s method or Runge–Kutta methods discretize the equation and compute the solution iteratively. However, these approaches require repeated calculations and may introduce computational latency. In applications that require real-time response, such as dynamic system modeling, control system analysis, and rapid prototyping, digital methods may suffer from inefficiency or delay. In contrast, analog circuits can implement differential equations directly using continuous-time signal processing. By representing system variables as voltage signals and constructing differentiator, scaling, and summation circuits with operational amplifiers, the mathematical relationships of differential equations can be implemented directly at the circuit level. For example, the proposed system can simulate the real-time response of dynamic systems such as mechanical vibration models or RLC circuit responses. In this approach, system variables such as displacement or current can be observed directly on an oscilloscope as time-varying signals. This provides a low-latency solution for dynamic system analysis, control system education, and rapid engineering prototyping. # Solution Overview This project aims to design and implement an analog differential equation solver using operational amplifier circuits. The system directly implements the mathematical relationships of a differential equation through analog signal processing, enabling continuous-time computation of dynamic system responses. In the proposed system, variables in the differential equation are represented as voltage signals. For instance, the system state $x(t)$ is represented by a voltage signal, while its time derivative $dx/dt$ is generated using an operational amplifier differentiator circuit. At the same time, weighted amplification and summation circuits are used to construct the algebraic terms on the right-hand side of the differential equation, such as coefficient multiplications and signal combinations. The entire circuit forms a feedback structure that ensures the signals within the system satisfy the constructed differential equation. When an input signal $u(t)$ is applied as an external excitation, the circuit generates the corresponding system response in real time. The resulting waveform can then be observed on an oscilloscope. Compared with conventional digital methods, this analog approach performs computation continuously without discretization or iterative numerical algorithms. As a result, the system can achieve low-latency computation and demonstrate the feasibility of using analog electronics for real-time dynamic system modeling. # Solution Components The system consists of several subsystems, each responsible for implementing a specific function required for solving the differential equation. ## Subsystem I: Differentiation Module - **Hardware I.a – Differentiator Circuit**: This module implements an operational amplifier differentiator circuit that computes the time derivative of an input signal. The circuit uses a capacitor–resistor network together with an op-amp to produce an output voltage proportional to the derivative of the input voltage. This module provides the fundamental operation required for representing derivative terms in the differential equation. - **Hardware I.b – Summation and Scaling Circuit**: This module uses operational amplifier summing amplifiers and resistor networks to implement weighted combinations of signals. By adjusting resistor values, the circuit can scale signals to represent coefficients in the differential equation. The circuit performs operations such as coefficient multiplication and signal addition, for example implementing expressions such as $ax$ or $ax+bu(t)$. ## Subsystem II: Input Signal Module - **Hardware II.a – Signal Generation**: This subsystem provides the external input signal $u(t)$ to the system. A function generator will be used to produce different types of excitation signals, such as step signals, sinusoidal signals, or square waves. These signals simulate external inputs to the dynamic system modeled by the differential equation. ## Subsystem III: Output Observation Module - **Hardware III.a – Oscilloscope Visualization**: The output voltage of the circuit represents the solution of the differential equation, corresponding to the system response over time. This signal will be connected to an oscilloscope, allowing real-time observation of system behavior such as oscillations, damping, or steady-state responses. ## Subsystem IV: Power Supply Module - **Hardware IV.a – Dual Power Supply**: Operational amplifiers require both positive and negative supply voltages to process signals that vary around zero. Therefore, the system will use a dual DC power supply providing approximately ±12 V to power the analog circuits. # Criterion for Success The project will be considered successful if the following criteria are satisfied: - **Accurate Differentiation**: The differentiator circuit must correctly compute the time derivative of the input signal and operate stably within the expected frequency range. - **Correct Equation Implementation**: The summation and scaling circuits must correctly implement the coefficients and mathematical structure of the target differential equation. - **Real-Time System Response**: When an excitation signal is applied, the system should produce a continuous output signal representing the system response in real time. - **Consistency with Theoretical Behavior**: The waveform displayed on the oscilloscope should match the expected theoretical behavior of the modeled differential equation, such as exponential decay, oscillatory motion, or steady-state response, within reasonable tolerance. |
||||||
| 43 | Cyber Guandan Tabletop Assistant with Real-Time Game Display and Event Monitoring |
Fan Zhang Wendao Yao Yushang Yang Zihan Zhou |
design_document1.pdf final_paper1.pdf other1.pdf |
Yushi Cheng | ||
| ## 1. Problem Definition and Motivation Physical card games such as Guandan are highly interactive and enjoyable, but they usually do not provide real-time information support for players or spectators. During a fast-paced game, it can be difficult to keep track of scores, tribute status, and recent game history. This problem becomes more obvious for new players, audiences, or demonstration settings where the game needs to be easier to follow. In addition, traditional tabletop games do not provide a convenient way to monitor unusual card events. If a card unexpectedly appears in or disappears from the active play region, it may be difficult to notice immediately or review afterward. To address this problem, this project aims to develop a vision-based Guandan tabletop assistant that can monitor a predefined play region and display useful game information in real time. By combining overhead vision with an assistant display, the system can help users better understand the game process while also providing basic event monitoring and replay capability. The success of this project will be evaluated based on the following criteria: - The system can detect card appearance or disappearance in a predefined tabletop region. - The system can display score, tribute status, and recent game history in real time. - The system can detect unexpected card events in the monitored region. - The system can provide an alert and a short replay clip when such an event is detected. - The system can operate as a stable tabletop demo with minimal manual setup. ## 2. Solution Overview The proposed solution integrates vision-based monitoring and real-time game display into a unified tabletop assistant system. An overhead camera captures the game area, and an embedded processing unit analyzes the monitored region to detect card changes and selected game events. Once a card event is detected, the system updates the assistant display with related game information such as scores, tribute reminders, and recent history. This allows both players and spectators to follow the game more easily. In addition, the system monitors the active play region for unexpected card appearance or disappearance events. When such an event occurs, the system issues an alert and provides a short replay clip for review. Compared with a normal physical card table, the proposed system adds real-time information support and event review while preserving the original gameplay experience. Compared with a more complex projection-based design, the proposed solution is more practical and easier to implement for a reliable classroom demonstration. ## 3. System Architecture and Components ### Vision Module The vision module captures the tabletop scene using an overhead camera and monitors a predefined play region. Its main function is to detect card appearance and disappearance events and provide visual input for later processing. ### Game Information Module This module maintains selected game information such as scores, tribute status, and recent play history. It updates the displayed information based on the recognized tabletop events. ### Assistant Display Module The assistant display module presents useful information on a separate screen instead of projecting directly onto the table. It shows scores, tribute status, recent history, system alerts, and replay output. ### Event Monitoring and Replay Module This module determines whether an unexpected card event has occurred in the monitored region. When such an event is detected, it generates an alert and saves a short replay clip for review. ### Embedded Control Module The embedded control module coordinates the camera, processing unit, and display subsystem. It is responsible for overall system operation and stable demo startup. ## 4. Criteria of Success Our project will be considered successful if it satisfies the following goals: - The system can correctly detect card appearance or disappearance events in the predefined play region under controlled lighting conditions. - The assistant display can update score, tribute status, and recent history with low visible delay. - The system can detect at least one or two predefined unexpected card event types and provide an alert within a short time. - The system can save and display a short replay clip of about 3 to 5 seconds for detected events. - The full system can run continuously for at least 20 minutes as a stable demonstration without major manual adjustment. |
||||||
| 44 | A World-Model based Infant Interaction Robot |
Ruijin Xu Zishuo Feng |
design_document2.pdf final_paper1.pdf presentation1.pptx video |
Gaoang Wang | ||
| # Problem Interactive robots designed for infant-oriented companionship, play, and sensorimotor stimulation have potential applications in safe and adaptive human-robot interaction. However, a key challenge is ensuring safe and robust interaction, as infant-like users or proxy moving objects may unpredictably grab, hit, or collide with the robot. Traditional reactive systems are often insufficient because they respond only after contact occurs, potentially too late to prevent harm to the robot or the infant. Current robotic systems lack the ability to anticipate infant-like actions and proactively adjust their behavior in real time, creating a safety gap in close-range human-robot interaction. # Solution Overview This project aims to develop a world-model-based intelligent mobile robot system that can operate safely and proactively in infant-like interaction scenarios. The system perceives nearby motion and approach behavior through onboard sensors such as ToF proximity sensors, bumper switches, IMU, and optionally a camera, utilizes a lightweight world model to predict short-term future states such as imminent contact or grab risk, and adjusts robot motion strategies in real time to enable dynamic and safe interaction. The emphasis is on predictive avoidance rather than reactive response, allowing the robot to anticipate and avoid potentially harmful situations before they occur. # Solution Components 1. Sensing Subsystem * Utilize ToF distance sensors, proximity sensors, bumper switches, and IMU to capture infant motion, approach patterns, and environmental context. * Detect approaching and possible grabbing-like behaviors in real time through multi-sensor fusion. 2. Embedded Processing & World Model * Run on an embedded platform (ESP32 or similar) for sensor data acquisition, timestamping, and real-time processing. * Implement a lightweight world model (small GRU/LSTM or 1D CNN) that learns latent dynamics from time-series sensor data and predicts future states (e.g., sensor values, grab risk) over a 0.5-1 second horizon. 3. Decision Logic & Robot Control * Use world model predictions to make decisions: if predicted future distance falls below a threshold, trigger proactive avoidance maneuvers. * Control robot chassis (2WD with motor driver) to execute behaviors such as retreating or changing direction. * Integrate basic interaction capabilities (speaker/buzzer for audio cues) for infant engagement. 4. Data Logging & Evaluation Pipeline * Implement data logging to synchronize sensor data with ground truth labels for model training and evaluation. * Enable comparison between reactive baseline (threshold-based) and prediction-based avoidance strategies. # Criterion for Success 1. The robot can collect stable real-time sensor data and move reliably on a mobile chassis. 2. A reactive baseline avoidance system is implemented and demonstrated. 3. A lightweight prediction model is trained and deployed for short-term risk prediction. 4. The robot can use model predictions to trigger proactive avoidance in real time. 5. In controlled tests, the prediction-based system demonstrates measurable improvement over the reactive baseline in at least one metric, such as lower contact rate, higher avoidance success rate, faster response, or greater maintained separation distance. |
||||||
| 45 | Intelligent Wearable Vision Systems for Assistive Perception |
Junchen He Mingyan Gao Shengnan Cai Yi Su |
final_paper1.pdf other1.pdf other2.pdf |
Wee-Liat Ong | ||
| # Request for Approval (RFA) ## ECE 445 / ME 470 – Spring 2026 **Project Title:** Intelligent Wearable Vision Systems for Assistive Perception ### Team Members | Name | UID | Major | | :--- | :--- | :--- | | Yi Su | 676182091 | Mechanical Engineering | | Mingyan Gao | 658581716 | Computer Engineering | | Shengnan Cai | 665630420 | Electrical Engineering | | Junchen He | 663319500 | Electrical Engineering | - **Date Submitted:** March 13, 2026 - **Course Instructors:** Prof. Weeliat Ong - **Suggested TA:** Zhao Ruolin (UID: 22571086) --- ## 1. Problem For visually impaired individuals, navigating everyday environments — hallways, crosswalks, stairs — requires either reliance on others or tools that fall significantly short of what the situation demands. Traditional aids like white canes provide limited spatial awareness, while existing smart glasses tend to either overwhelm users with indiscriminate scene description or fail to operate reliably in real-world conditions. A key shortcoming of current systems is that they treat perception as a static problem: they do not adapt to whether the user is walking briskly through a crowd, pausing at a curb, or turning into an unfamiliar corridor. The result is feedback that arrives too late, too often, or without meaningful prioritization — reducing rather than enhancing the user's sense of control. ## 2. Solution Overview We propose a wearable vision system — designed to be worn as glasses or integrated into a lightweight cap — that uses on-device computer vision to continuously monitor the environment and relay only the information most relevant to safe navigation. Rather than describing everything the camera sees, the system focuses on hazards that require near-term action: obstacles at ground level, steps, approaching pedestrians, and doorways. Feedback is delivered through bone-conduction audio and small vibration motors, keeping the user's hands free and their ears open to surrounding sound. A distinguishing feature of the design is that the system monitors the user's own motion through an inertial sensor and adjusts its behavior accordingly — more alert and faster to flag hazards when the user is moving, quieter and less intrusive when they have stopped. This context-sensitivity is what we believe makes the difference between a system that genuinely aids navigation and one that simply adds noise. ## 3. Components The system is organized into four subsystems, all housed within a wearable form factor. ### 3.1 Sensing Subsystem A compact RGB or RGB-D camera captures the scene ahead, while an IMU (accelerometer and gyroscope) tracks the user's movement and orientation. The camera will be selected based on trade-offs between power draw, weight, and depth sensing quality; we are currently evaluating a few candidates including OV-series modules and Intel RealSense D4xx compact variants. ### 3.2 Processing and Intelligence Subsystem The core computation runs on a small edge board — likely a Raspberry Pi 5 or NVIDIA Jetson Nano depending on the latency and power budget we settle on during prototyping. A lightweight object detection model (MobileNet-SSD or similar) handles hazard classification in real time. A separate logic layer fuses the IMU data to decide when and how urgently to trigger feedback, and estimates rough distances to flagged objects using depth information or monocular depth cues. ### 3.3 Feedback Subsystem Audio output uses a bone-conduction transducer so the user can hear ambient sound simultaneously. Haptic output comes from small eccentric rotating mass (ERM) motors positioned to give a rough directional sense — for instance, left versus right — when a hazard is detected nearby. The feedback modality, timing, and phrasing will be iterated based on informal usability tests during development. ### 3.4 Power and Mechanical Subsystem Power is provided by a rechargeable Li-Po cell sized to last a full day of use. The enclosure will be designed with wearability as a primary constraint — lightweight materials, balanced weight distribution, and enough environmental sealing to be usable outdoors in light rain or dust. ## 4. Criteria of Success We consider the project successful if the system demonstrates reliable and timely hazard perception in realistic navigation scenarios, both indoors (hallways, stairwells) and outdoors (sidewalks, crosswalks). We will evaluate the following outcomes through structured tests with participants. - **Perception reliability:** The system should detect and correctly identify the target hazard categories (pedestrians, vehicles, curbs, stairs, doors) with high consistency across typical indoor and outdoor lighting conditions, maintaining a low enough false alarm rate that the feedback remains trustworthy rather than distracting. - **Response timeliness:** End-to-end latency — from camera capture to delivered feedback — should be short enough that a walking user has sufficient time to react and adjust their path. The system should feel responsive in everyday use rather than lagged. - **Wearability:** The assembled system should be light and compact enough for comfortable extended wear, with battery life sufficient to cover a full day of normal use without recharging. - **Usability:** In informal blindfolded navigation tests, participants unfamiliar with the system should be able to interpret the feedback cues quickly and respond appropriately to introduced hazards without verbal instruction. We will iterate on the feedback design until this is achieved to a satisfactory degree. |
||||||
| 46 | Voice-Controlled Robotic Study Assistant |
Jiaxuan He Qi Jin Shuohan Fang Yicheng Chen |
design_document1.pdf final_paper1.pdf final_paper2.pdf other1.pdf proposal1.pdf video |
Hua Chen | ||
| # Problem According to the newest data of the World Health Organization (WHO), there are about 295 million people who suffer from vision impairment. And 43 million people among them are blind. Moreover, the number of people with hand disabilities is from tens of millions to hundreds of millions. With the development of assistive technologies, the quality of life of these groups of people is improving. However, there’s still a lack of an all-in-one solution. Existing page-turning devices often provide not enough interactive ways, such as keyword search and turning pages by voice command. Based on this observation, we aim to develop an auto page-turning machine with voice control and text recognition, which truly provides aid to millions of people with disabilities who are currently marginalized by traditional print media. # Solution Overview Our solution is a voice-controlled, autonomous reading machine that combines mechanical page manipulation with computer vision. A central microcontroller manages the physical navigation of the book. To turn a page, the system coordinates three steps: two actuated paperweights secure the book, a robotic arm with a vacuum suction cup vertically lifts the top sheet, and a motorized swing arm sweeps the page across the binding. This design allows for hands-free, bidirectional page turning. To ensure mechanical reliability, we paired this physical setup with a fixed overhead camera and a computer vision pipeline. Because thick textbooks naturally curve near the binding, our software first applies a geometric dewarping algorithm to digitally flatten the captured images. This improves the accuracy of the Optical Character Recognition (OCR) engine. The camera also forms a closed-loop control system by reading page numbers after every turn. If the system detects that multiple pages were accidentally grabbed, the microcontroller automatically triggers a reverse-sweep recovery sequence to fix the error. Users control the entire system using natural language voice commands via a microphone. They can request absolute or relative page navigation, such as "turn to page 45" or "go back one page." Once the system verifies it has reached the correct page, it waits for further instructions. If the user commands it to read, the system uses the extracted OCR text to provide text-to-speech (TTS) playback through an integrated speaker. Additionally, the software features a voice-activated keyword search. Users can ask the machine to locate specific terms on the open spread, and the system will verbally identify the exact paragraph or read the relevant context, providing a fully interactive study experience. # Solution Components The proposed system consists of several integrated components that enable voice-controlled interaction with physical reading materials. These components work together to assist users, particularly individuals with limited upper-limb mobility, in accessing and navigating printed documents independently. **1. Voice Command Interface** This module allows users to control the system using predefined voice commands. The system recognizes commands such as “next page,” “previous page,” “read page,” “book one,” and “book two.” These commands enable the user to navigate through physical reading materials without using their hands. The recognized speech input is processed and translated into control signals that trigger the corresponding system actions. **2. Robotic Page-Turning Mechanism** This component performs the physical manipulation of paper pages. A robotic mechanism, consisting of actuators and a page-lifting structure, is designed to lift and flip individual pages of a book or document. The mechanism must operate carefully to avoid tearing or damaging the paper while ensuring that only a single page is turned at a time. The system is designed to handle common book and document sizes within a specified range. **3. Dual Document Workstations** The system includes two predefined workstations where different physical reading materials can be placed before operation begins. These workstations allow the user to switch between two separate documents using voice commands. For example, one workstation may contain a bound textbook while the other contains stapled lecture notes. This feature allows users to interact with multiple learning materials without manual assistance. **4. Vision-Based Page Monitoring** A camera system continuously monitors the document during operation. This module captures images of the current page and uses computer vision techniques to detect whether a page-turning action has been successfully completed. It can also identify possible errors such as incomplete page flips, page misalignment, or paper sticking. The visual feedback helps improve system reliability and provides useful information for system control and debugging. **5. Text Recognition and Audio Reading Module** Using Optical Character Recognition (OCR), this module extracts textual content from the captured page images. The recognized text is then processed by a text-to-speech (TTS) system that reads the content aloud to the user. This function allows visually impaired users or users who prefer auditory feedback to access the information on the current page without needing to read the physical text directly. **6. System Control and Integration** This module serves as the central controller of the system. It coordinates all components, including voice input processing, robotic page-turning actions, vision feedback, and audio output. The control module ensures that commands are executed in the correct sequence and that feedback from sensors and the vision system is used to verify successful operations. This integration allows the system to function reliably as a unified assistive reading platform. # Criterion for Success 1. The system shall correctly recognize and execute predefined voice commands such as “next page,” “previous page,” “turn to page X,” “read page,” “book one,” and “book two” in at least 8 out of 10 trials, with each action beginning within 3 seconds in a quiet indoor environment. 2. The system shall autonomously switch between two predefined document workstations, one containing a bound textbook and one containing stapled lecture notes, without requiring manual document replacement during operation, in at least 8 out of 10 trials. 3. For each of the two document types, the system shall successfully turn a single page forward and backward in at least 8 out of 10 trials, without damaging the paper, unintentionally turning multiple pages, or causing visible permanent damage. 4. The vision system shall correctly detect whether a page-turning action has succeeded or failed and identify common errors such as incomplete flips, page misalignment, or multiple-page pickup in at least 8 out of 10 trials; when an error is detected, the control system shall initiate a corrective action. 5. For a predefined set of printed textbook and lecture-note pages, the OCR and audio module shall extract the main textual content and read it aloud with at least 90% text accuracy, and the keyword search function shall correctly identify whether a requested keyword appears on the current page in at least 8 out of 10 trials. |
||||||
| 47 | Design of a Mechatronic Physical Road-Crossing Game System |
Tianxi Zhu Yuxuan Liu Zhuo Li Zihao Wu |
design_document1.pdf final_paper1.pdf final_paper2.pdf |
|||
| # Problem Most games today exist only in digital environments such as computers or mobile devices. While these games are entertaining, they do not demonstrate how mechanical systems, electronic circuits, sensors, and control systems interact in real physical systems. For students studying mechatronics and embedded systems, building a physical interactive system provides a better understanding of how hardware and control systems work together. The challenge is to design a system that coordinates multiple moving vehicles, detects collisions, and responds to player inputs while maintaining stable operation. The system should also be mechanically simple and reliable enough to operate repeatedly during demonstrations. To address this challenge, we propose a physical road-crossing game system in which a player-controlled vehicle attempts to cross a simulated road while avoiding moving traffic vehicles. # Solution Overview The proposed system is a physical game platform that integrates mechanical structures, electronic circuits, sensors, and a microcontroller control system. The platform represents a road environment. Several traffic vehicles move across the road along fixed tracks using motor-driven mechanisms. These vehicles simulate moving traffic and act as obstacles. The player controls a small vehicle that attempts to cross the road from a starting position to a goal area. During the game, sensors monitor the positions of vehicles and detect collisions. If the player vehicle collides with a traffic vehicle, the system registers a failure condition. If the player reaches the goal safely, the system registers a successful crossing. A microcontroller coordinates all system operations including motor control, sensor monitoring, and game logic. # Solution Components The system is composed of several subsystems that together implement the road-crossing game. ## Subsystem I: Player Vehicle Control - **Hardware I.a – Motor Drive** A DC motor or servo motor drives the movement of the player vehicle. - **Hardware I.b – User Input Interface** Buttons or simple control inputs allow the player to control the vehicle. ## Subsystem II: Traffic Vehicle Motion - **Hardware II.a – Motorized Traffic Vehicles** Multiple vehicles move along fixed tracks across the road using DC motors. - **Hardware II.b – Mechanical Track Structure** A rail or track guides the motion of the traffic vehicles. ## Subsystem III: Collision Detection - **Hardware III.a – Collision Sensors** Sensors such as infrared sensors or contact switches detect collisions between vehicles. ## Subsystem IV: System Control - **Hardware IV.a – Microcontroller Controller** A microcontroller manages motor control, sensor input, and overall game logic. # Criterion for Success - Traffic vehicles move smoothly along their tracks in at least **8 out of 10 trials** without mechanical failure. - The player vehicle responds correctly to user inputs and can cross the road. - The collision detection system correctly detects collisions between vehicles. - The system correctly determines game success or failure. - All mechanical, electronic, and control subsystems operate together as a complete mechatronic system. |
||||||
| 48 | Intelligent Net-Energy Optimization System for Distributed Photovoltaic Nodes in Microgrids |
Minghao Fang Yifei Liu Yikai Zhang Ziru Niu |
design_document1.pdf final_paper1.pdf other1.pdf |
Ruisheng Diao | ||
| Problem In modern microgrids, Distributed Energy Resources (DERs), particularly small-scale photovoltaic systems, suffer from significant efficiency losses due to the misalignment between solar panels and the sun. While dual-axis tracking systems exist, traditional active tracking methods often consume more power in actuation (motors) than they gain in generation, especially during intermittent cloud cover or low-light conditions. There is a lack of low-cost, adaptive control strategies that can autonomously evaluate the "net energy gain"—balancing the energy cost of moving against the potential generation revenue—in real-time. Solution Overview This project seeks to develop an intelligent, edge-computing-based control solution that maximizes the net energy yield of a PV node using accessible, cost-effective hardware, ensuring economic viability for small-scale microgrid applications. Our system will utilize a Master-Slave architecture, integrating a Raspberry Pi for high-level computing (e.g., net-energy optimization algorithms) and a microcontroller (e.g., STM32) for hard-real-time motor execution. It will feature one-button autonomous calibration and real-time visualization of energy data. Solution Components Software Component: Edge-computing logic (e.g., Q-Learning or threshold-based algorithms) on the Raspberry Pi to decide optimal tracking strategies based on real-time irradiance and motor power consumption. Real-time embedded control code on the microcontroller for accurate sensor polling, PWM generation for motors, and serial communication with the Raspberry Pi. Data visualization software to drive an OLED screen, displaying current voltage, net power gain, and AI status. Hardware Component: A custom-designed PCB integrating robust power management (essential for simultaneously powering the Raspberry Pi and motors via battery/PV), stepper/servo motor driver circuits, and sensor interfaces. Microcontroller (e.g., STM32/ESP32) and Raspberry Pi boards. Sensor array: Current/Voltage sensors (e.g., INA219) for power calculation, and photoresistors/LDRs for light tracking. Dual-axis pan-tilt mechanical structure, solar panel, and a stable chassis. Criteria of Success The system initiates self-calibration and begins autonomous tracking immediately upon a single button press, requiring no external computer connection. The system successfully tracks the brightest light source under normal conditions. The adaptive control algorithm successfully pauses motor actuation during simulated low-light or rapidly fluctuating light conditions, demonstrating an avoidance of negative net energy gain compared to a continuous tracking baseline. The OLED display accurately shows real-time system metrics (voltage, current, power status). Distribution of Work Ziru, Niu (EE) & Yifei, Liu (ECE): Responsible for the custom PCB design, power management circuitry, hardware sensor integration, and underlying microcontroller programming for motor control and data acquisition. Minghao, Fang (ECE): Responsible for developing the edge-computing optimization algorithms on the Raspberry Pi, serial communication protocols, and the OLED data visualization software. Yikai, Zhang (ME): Responsible for the physical design and fabrication of the dual-axis pan-tilt mechanism, ensuring the structural stability of the chassis, and managing the heat dissipation and mounting of the electronic components. |
||||||
| 49 | Eco-Trim: Smart Scratch Pad Recovery System |
Lehan Pan Tianyi Xu Zhizheng Ju Zihan Wang |
design_document1.pdf final_paper1.docx final_paper2.docx proposal1.pdf |
|||
| **Team Members:** Zhizheng Ju (zju4), Tianyi Xu (tianyi25), Lehan Pan (lehanp2), Zihan Wang (zihan20) ### Project Description **Problem** University libraries generate massive paper waste from partially used or single-sided prints. This innovation aims to reclaim these blank spaces locally, avoiding resource-heavy chemical recycling. The system must autonomously feed discarded stacks, identify blank areas, and extract usable scratch pads. **Solution Overview** Eco-Trim is an automated mechatronic system. Unlike standard shredders that destroy paper, or industrial recyclers that require wet pulping, our dry, localized approach actively salvages usable sections. The complexity lies in synchronizing optical detection with mechanical actuation, which is highly feasible given our interdisciplinary team. An ECE-driven solution is absolutely necessary because the system requires a custom PCB, a microcontroller running a finite state machine (FSM), and opto-isolated motor drivers to process sensor data and execute high-torque cutting safely.The entire system operates with a single push of a button, strictly adhering to the one-button activation rule. **Components** * **Mechanical Feeding:** Motorized friction rollers engineered to ensure reliable single-sheet feeding and flattening without jamming. * **Optical Detection:** A custom PCB integrating an MCU and optical contrast sensors to scan and map printed versus blank areas. * **Cutting Actuation:** A stepper-motor-driven blade mechanism designed for precise, automated trimming based on sensor feedback. * **Safety Interlocks (Critical):** Essential IR light curtains and enclosure limit switches that instantly sever high-voltage motor power upon detecting human proximity, preventing injury and ensuring safe public usability. **Criteria of Success** 1. System continuously feeds at least 5 single sheets of paper without mechanical jamming. 2. Optical sensors correctly differentiate printed from blank areas, triggering the appropriate cutting logic. 3. The actuation blade cleanly severs the paper into uniform pads without tearing. 4. Hardware safety interlocks physically disconnect motor power within 100ms when the enclosure is opened or the light curtain is breached. |
||||||
| 50 | Automatic Sorting Robotic Arm for Table Tennis Balls |
Siqi Pan Xucheng Wu Zhonghao Wang |
design_document2.pdf final_paper1.pdf other1.docx other2.pdf presentation1.pptx proposal1.pdf video1.mp4 |
|||
| # Team Members - Junway Lin (3220300390) - Siqi Pan (3220111242) - Xucheng Wu (3220111011) - Zhonghao Wang (3220111411) # Problem Table tennis balls exhibit varying sizes, weight, color, material composition, and surface markings due to evolving manufacturing standards and regulations. The current ITTF ruling specifies that a standard ball legal for play must meet the following requirements: - 40 mm in diameter - 2.7 g in weight - ABS plastic construction - White or orange matte appearance Prior to the regulation changes in October 2000, table tennis balls legal for play were: - 38 mm in diameter - 2.5 g in weight - Made of celluloid, a highly flammable and lightweight synthetic plastic These changes were implemented to slow down the game and improve visibility for spectators and television audiences. Notably, some differences among table tennis balls are visually difficult to distinguish, which complicates manual sorting. In training, storage, and equipment management contexts, these variations can lead to inefficiencies and classification errors when sorting is performed manually. # Solution Overview This project addresses the problem by designing and developing an automated robotic system capable of accurately identifying and sorting table tennis balls based on their physical and visual characteristics. The system will measure key attributes of each ball, including: - Diameter - Weight - Color - Logo presence These measurements are processed by an embedded controller to classify each ball and actuate a robotic arm to place it into the appropriate category. The system architecture combines: - Mechanical design - Multimodal sensing - Embedded processing - Controlled actuation Compared to manual sorting, the proposed solution improves accuracy, consistency, and efficiency while reducing human effort. In contrast to vision-only systems, the use of multiple sensing modalities enables more robust and reliable classification, particularly in cases where visual ambiguity or environmental variability may degrade performance. # Solution Components ## Input Box (Passive Storage) - **Geometry:** Sloped walls or funnel shape to keep balls clustered and accessible - **Surface:** Low-friction lining to prevent jamming - **Availability Detection (for robustness):** - Simple IR break-beam or ToF sensor aimed at the pile to corroborate “empty” detection from vision ## Sensing Platform (Measurement + Fixturing) - **Mechanical stabilization:** - Concave cup or V-groove with known geometry that allows for ease of ball placement and pickup and ensures the ball is fixed in place - **Diameter measurement:** - ToF sensor (e.g., VL53L0X) with fixed reference geometry - **Presence detection:** - IR break-beam or reflective IR sensor to trigger measurement - **Design note:** - Platform height and geometry should be consistent with the arm’s pickup kinematics ## Robotic Arm with Integrated Vision (Feeding + Inspection + Sorting) - **Functions combined:** 1. Pick ball from input box 2. Place ball on sensing platform 3. Capture image for color/logo 4. Pick ball from platform 5. Place into correct output bin - **Actuation:** - 3-4 DOF servo-based arm (PWM-controlled servos) - Actual model of arm undetermined - **End effector (critical choice):** - **Suction gripper:** - Small vacuum pump + nozzle - High reliability for smooth, lightweight balls - Might be unreliable if the balls are not clean - **Alternative:** - Compliant finger gripper with rubber padding - **Vision module:** - Camera (e.g., Raspberry Pi Camera) mounted near the end effector - **Vision tasks:** - Color classification (RGB thresholding) - Logo detection (contrast/contour-based) - **Key implications of using arm for feeding:** - Requires pose consistency of balls in input box - May need simple bin shaping - Increases cycle time, since feeding is no longer parallelized - Eliminates the need for feeder actuators, reducing mechanical complexity ## Output Boxes (Sorting Bins) - Fixed bin positions mapped to arm coordinates - Sized to tolerate placement error ## Custom PCB Controller (Centralized Control + Interfaces) ### Core Architecture - **Microcontroller (primary):** - STM32 (e.g., STM32F4 series) for real-time control and integration - **Vision processing (recommended split architecture):** - Raspberry Pi handles image processing - Communicates with PCB via UART/SPI ### PCB-Integrated Subsystems - **Sensor interfaces:** - HX711 circuit for load cell (can be integrated or provided as module footprint) - I2C buses for ToF sensors - GPIO for IR sensors - **Motor control:** - PWM outputs for servo control (arm + optional actuators) - Power drivers and proper current routing for motors - **Power management:** - Separate regulated rails: - 5-6 V high-current rail for servos - 3.3 V for logic and sensors - Decoupling and filtering - **Communication:** - UART/SPI link to vision processor - **Protection and layout considerations:** - Ground plane separation (analog vs. digital) - Noise isolation for weight measurement accuracy # Criterion for Success ## Ball Handling and Manipulation - The robotic arm shall successfully pick up a table tennis ball from the input box and place it on the sensing platform with a success rate of **≥ 90%** over **20 consecutive trials**. - The arm shall transfer a classified ball from the sensing platform to the correct output bin with a placement accuracy of **±2 cm** from the target bin center in at least **90% of trials**. ## Sensing Accuracy - The sensing platform shall measure ball weight with an error of **≤ ±0.02 g** compared to a calibrated reference scale. - The diameter estimation system shall classify balls as **38 mm vs 40 mm** with **≥ 95% accuracy** under controlled positioning conditions. - The vision system shall correctly classify ball color (**white vs orange**) with **≥ 95% accuracy** under consistent lighting conditions. - Logo presence detection shall achieve **≥ 85% accuracy**, acknowledging higher variability. ## System Reliability - The full system shall successfully complete at least **30 consecutive sorting cycles** with no more than **2 total classification or actuation failures**. - The system shall recover from a failed grasp or missed detection without manual reset in **≥ 80% of failure cases**. |
||||||
| 51 | PHOTOVOLTAIC POWER GENERATION CHARGER |
Guangjun Xu Sunhao Zhang Xu Li |
design_document1.pdf final_paper1.pdf final_paper2.pdf |
|||
| # PHOTOVOLTAIC POWER GENERATION CHARGER ## 1. PROBLEM DEFINITION AND MOTIVATION With the increasing demand for clean and sustainable energy, photovoltaic power generation has become an important solution for reducing dependence on conventional fossil fuels. However, in many daily and small-scale applications, electrical devices still rely heavily on grid power or disposable batteries, which may increase energy costs and create environmental burdens. This project aims to develop a photovoltaic power generation charger that can convert solar energy into electrical energy and use it to charge electronic devices or rechargeable batteries. The system focuses on collecting solar energy through photovoltaic panels, regulating the output power, and delivering stable charging performance under different light conditions. The project demonstrates a complete renewable-energy-based charging process, from solar energy collection to electrical energy conversion and battery charging. Its success will be evaluated based on whether the system can efficiently harvest solar energy, provide stable voltage and current output, and charge target devices safely and reliably with minimal external power support. --- ## 2. SOLUTION OVERVIEW The proposed solution integrates solar energy harvesting, power regulation, and battery charging into a unified charging workflow. A photovoltaic panel captures sunlight and converts it into electrical energy. Since the output of the panel may vary depending on sunlight intensity, a power management circuit is used to regulate the generated energy and provide a stable electrical output. After regulation, the charging system delivers suitable voltage and current to the target load, such as a rechargeable battery or a low-power electronic device. The system may also include a monitoring function to display charging status, output voltage, current level, or battery condition. In this way, the project performs both energy conversion and device charging as a complete renewable power application. The feasibility of the system is supported by the availability of standard hardware components such as photovoltaic panels, charge controllers, voltage regulators, batteries, and monitoring modules, as well as mature circuit design methods for energy conversion and charging control. --- ## 3. SYSTEM ARCHITECTURE AND COMPONENTS ### PHOTOVOLTAIC ENERGY COLLECTION MODULE The photovoltaic energy collection module uses a solar panel to capture sunlight and convert it into electrical energy. The output power depends on environmental conditions such as light intensity and panel orientation. This module serves as the primary energy source of the system. ### POWER REGULATION AND CONTROL MODULE This module receives the electrical energy generated by the photovoltaic panel and regulates it into a stable and usable form. It may include voltage regulation, current control, and protection functions to ensure safe and efficient charging. It coordinates the overall energy flow and maintains proper operating conditions for the charger. ### ENERGY STORAGE AND CHARGING MODULE The energy storage and charging module is responsible for storing electrical energy in a rechargeable battery or directly charging an external device. Based on the regulated output, the module manages the charging process to improve charging efficiency and protect the battery or load from overcharging or unstable input power. ### MONITORING AND OUTPUT MODULE After power conversion and charging, the system provides the final output to the target device or storage unit. This module may also display system information such as solar input condition, charging status, battery level, or output voltage and current. It completes the final delivery and user interaction step of the photovoltaic charging system. |
||||||