Project
| # | Title | Team Members | TA | Documents | Sponsor |
|---|---|---|---|---|---|
| 75 | RailRider (Reaction-Wheel Uni-Wheel Inspection Robot with Vision) |
James Recera Varun Sharma Zhanshuo Zhang |
Abdullah Alawad | design_document1.pdf design_document2.pdf final_paper1.pdf proposal1.pdf video |
|
| # Title RailRider (Reaction-Wheel Uni-Wheel Inspection Robot with Vision) # Team Members - Zhanshuo Zhang (zz128) - Varun Sharma (varuns10) - James Recera (jrecera2) # Problem A lot of important inspection locations are basically “thin-structure environments” where a normal robot is awkward or unsafe: narrow beams, cable trays, ladder racks, pipe-rack edges, and long tunnel-like spaces. These places show up in real settings like data centers (overhead cable management and airflow issues), HVAC/ventilation runs (debris, blockages, moisture), industrial facilities (leaks/labels/fasteners), and even “space-inspired” scenarios like a lunar/martian tunnel scout where falling off an edge or getting stuck could mean mission failure. A typical RC car is too wide and needs turning radius, and a drone is loud, short battery life, and often not allowed indoors (plus it struggles in confined, dusty, GPS-denied spaces). We want a compact platform that can move on narrow structures and produce useful inspection results instead of only streaming video. # Solution We will build a reaction-wheel stabilized uni-wheel robot that can travel along a narrow beam/rail while carrying a camera-based perception payload. The core idea is that the robot can balance itself with a tiny contact footprint, so it can ride on structures that would make a 4-wheel robot fall off. Our robot will support two main modes: 1. Teleop + safety override: user drives it, but the robot prevents unsafe motions near edges/obstacles. 2. Assisted inspection: the robot follows a beam/rail direction using perception cues and logs simple “inspection events” (marker reached, obstacle detected, possible defect). The perception system could detecting drop-offs or obstacles early enough to stop, and flagging inspection targets for different missions (some object detection and segmentation maybe using ai). # Solution Components **Subsystem 1: Main Control PCB** Custom PCB that does: power distribution, motor control, sensor IO, and communication. - MCU: ESP32-S3 (WiFi/BLE + good performance for control + telemetry) - IMU: BNO055 (easier) or ICM-20948 (harder but common) for orientation feedback - Reaction wheel motor driver: 3-phase BLDC driver stage (selected based on motor choice; goal is closed-loop reaction wheel torque control) - Wheel motor driver: brushed DC driver or BLDC driver depending on drivetrain - Power rails: battery → buck converters (example: 2S LiPo to regulated 5V + 3.3V) - Current/voltage sensing: measure battery + motor current for safety cutoff / stall detection - Connectors: I2C header for ToF/thermal, UART/USB header for perception module, debug header, kill switch - Safety + reliability: heartbeat/watchdog input from the CV module so the MCU can default to “safe stop” if perception freezes **Subsystem 2: Balancing and Moving** This subsystem keeps the robot upright and moves it forward. - Reaction wheel assembly: BLDC motor + flywheel disk (hub + added rim mass for inertia) - Drive wheel: geared motor or hub motor depending on size and torque needs - Control loop: IMU → controller → reaction wheel torque (and wheel torque as needed) **Subsystem 3: CV Perception Payload (camera or optional radar)** - Forward-facing camera for obstacle detection and logging markers/labels for missions - Onboard lighting (LED ring/light bar) for dark environments - Multizone ToF mounted as a hard safety override, when encounter sudden gaps or obstacles - (Optional) Thermal array (e.g., MLX90640) to flag hotspots - Possibly replace the camera with radar **Subsystem 4: Communications with users** - WiFi video/telemetry stream (ESP32 + CV module stream) - Simple laptop dashboard: live video, distance/edge warnings, “event log” (marker reached, obstacle, stop triggered) **Subsystem 5: Mechanical Structure** - Protective cage so if it tips it doesn’t destroy the camera - Modular mounting plate for sensors # Criteria For Success 1. Balance: robot can self-balance in place for ≥ 60 seconds without external support. 2. Narrow-structure traversal: robot can traverse a 2 m rail/beam (target width chosen for our demo rig) at slow speed without falling off. 3. Safety override: perception-based override stops the robot before a drop/obstacle with ≤ 20 cm stopping distance at test speed. 4. Inspection output: robot produces a structured event log with 3 event types, for example: “marker reached / tag detected,”; “obstacle detected / stop triggered,”; “possible anomaly (debris/loose cable) flagged,”; (optional) “thermal hotspot flagged.” # References: (for future project implementation) (1) “The Wheelbot: A Jumping Reaction Wheel Unicycle” (IEEE Robotics and Automation Letters, Vol. 7, No. 4, pp. 9683–9690, Oct. 2022). (2) https://github.com/peng-zhihui/ONE-Robot |
|||||