# Title Team Members TA Documents Sponsor
10 Smart Laundry FoldBot
Channing Liu
Jiadong Hong
Jialin Shang
Weijie Liang
Yu Lin
# Smart Laundry FoldBot RFA

## Team Member

Jiadong Hong EE
Qianqi Liu ME
Jialin Shang CompE
Weijie Liang CompE

## Problem

Laundry folding, a seemingly mundane task, can be surprisingly time-consuming, tedious, and even physically demanding. This project aims to enhance our overall well-being and quality of life by addressing the challenges associated with this commonplace but often underestimated activity. By alleviating the burden of laundry folding, the system we propose aims to liberate individuals to focus on more meaningful pursuits, contributing to a more harmonious and productive home environment.

The primary challenge lies in developing a sophisticated machine capable of efficiently automating the clothing recognition and folding process. The system should integrate advanced computer vision capabilities to accurately identify and categorize different types of clothing items, such as shirts, pants, dresses, and more. Moreover, it must be adaptable to varying sizes and clothing styles, ensuring the folding process accommodates the diverse range of garments found in typical households.

## Solution Overview

Our system automates laundry folding through:

**Core Boards:** Four motorized boards fold clothing sequentially—Left, Right, Lower, and Upper—for precision. The Upper Board aids in easy clothing removal.

**Expansion Plates:** Three adjustable plates adapt to clothing sizes, ensuring comprehensive folding for different dimensions.

**CV Assistance:** We would use advanced computer vision for accurate clothing recognition and spatial understanding.

**Kinetic Control System:** We would employ Reinforcement Learning for optimal folding and Exception-handling Algorithms for real-time adaptation.

Our Automated Clothing Recognition and Folding System integrates these components, providing an efficient and user-friendly solution for a more harmonious and productive home environment.

## Solution Components

### Core Boards

This component is essentially the primary folding mechanism, consisting of four specialized boards, each powered by an electric motor. These boards are designed to fold 180 degrees, enabling the sequential folding of clothing placed on them. The four boards are:

#### a. Left Core Board:

\- Positioned on the left side.

\- Folds 180 degrees to the right.

\- This action folds the left portion of the clothing (e.g., the left side of a shirt).

#### b. Right Core Board:

\- Located on the right side.

\- Folds 180 degrees to the left.

\- This mirrors the left core board's action, folding the right portion of the clothing.

#### c. Center Lower Core Board:

\- Situated below the central part of the clothing.

\- Folds upwards 180 degrees.

\- This folding step works on the lower part of the clothing, bringing it upwards and typically folding the garment in half.

#### d. Center Upper Core Board:

\- Located above the central part of the clothing.

\- Also folds upwards 180 degrees.

\- Completes the folding process by folding the upper portion of the garment. At this stage, the clothes are fully folded.

\- This board may interact with an external system, such as a conveyor belt, to move the folded clothing away from the machine.

### Expansion Plates

This component provides the system with the flexibility to handle various sizes and types of clothing. It comprises three adjustable plates:

#### a. Left Expansion Plate:

\- Adjacent to the left core board.

\- Capable of extending or retracting to accommodate different clothing sizes.

\- Specifically, it adjusts for clothing parts that extend beyond the left core board, like long sleeves, folding them appropriately.

#### b. Right Expansion Plate:

\- Positioned next to the right core board.

\- Functions similarly to the left expansion plate but on the right side.

\- Adjusts for the parts of the clothing that exceed the right core board.

#### c. Lower Expansion Plate:

\- Located below the central lower core board.

\- Operates under the same principle as the other expansion plates.

\- Adjusts for clothing parts that extend beyond the central lower core board, ensuring a complete and neat fold.

### CV Assistance

#### **Object Detection:**

Utilize sophisticated object detection algorithms, notably YOLO (You Only Look Once) or Faster R-CNN, to discern the spatial coordinates and categorical attributes of clothing articles. This facilitates a nuanced understanding of the depicted garments.

#### **Image Segmentation:**

Apply cutting-edge image segmentation methodologies, exemplified by Mask R-CNN or SAM, to differentiate various clothing items. This process effectively isolates clothing articles from the background, providing clear delineations that contribute to a detailed understanding of their spatial relationships and visual attributes.

### Kinetic Control System

#### **Optimization Algorithms:**

Reinforcement Learning: Adopt methodologies rooted in reinforcement learning paradigms, including Deep Reinforcement Learning (DRL), to facilitate the acquisition of optimal folding strategies through iterative learning mechanisms.

#### **Exception Handling Algorithms:**

Model Predictive Control (MPC): Implement MPC strategies for real-time adaptation of robotic arm dynamics, ensuring the accommodation of anomalous scenarios during the unfolding intricacies of clothing folding.

Sliding Mode Control: Harness the robust attributes of sliding mode control mechanisms to mitigate uncertainties and adapt to dynamic variations encountered during the operational course.

## Criterion for Success

The success of the Automated Clothing Recognition and Folding System will be measured based on the achievement of the following key criteria:

**Precision in Folding:** The system must consistently fold various types of clothing items with a high degree of precision, resulting in neatly organized garments.

**Integration of CV and Kinetic Control:** The successful integration of computer vision techniques for accurate clothing recognition (CV Assistance) and kinetic control algorithms (Kinetic Control System) to achieve optimal folding strategies.

**User-Friendly Interface:** The interface must be intuitive and user-friendly, allowing users to interact easily with the system and monitor the folding process.

**Safety:** Implementation of safety features is crucial to prevent accidents or damage to clothing items, ensuring a secure and risk-free operation.

ML-based Weather Forecast on Raspberry Pi

Xuanyu Chen, Zheyu Fu, Zhenting Qi, Chenzhi Yuan

Featured Project

#Team Members

Zheyu Fu ( 3190110355)

Xuanyu Chen ( 3190112156)

Chenzhi Yuan ( 3190110852)

Zhenting Qi ( 3190112155)


Weather forecasting is crucial in our daily lives. It allows us to make proper plans and get prepared for extreme conditions in advance. However, meteorologists always get it wrong half of the time and still keep their job :) To overcome the limitations of traditional weather forecasting, machine learning models have become increasingly important in weather forecasting. Building our own weather forecast ML system is a perfect idea for us to analyze vast amounts of area data and generate more accurate and timely weather predictions on the go in our surrounding areas.

#Solution Overview

A weather forecast system can be created by using a few different hardware components and software tools. Our solution mainly consists of two parts. For weather measurement and data collection, temperature, humidity, and barometric pressure sensors are considered the main components. A machine learning-based algorithm is to be applied for data analysis and weather predictions.

#Solution Components

##Hardware Subsystem

Due to the complexity of weather conditions, our system incorporates the following weather indicators and their corresponding collectors:

-a barometric pressure sensor, a temperature sensor, and a humidity sensor

-a digital thermal probe for heat distribution

-an anemometer for wind speed, wind vane for wind direction, and rain gauge for precipitation

The aforementioned equipment would be integrated into a single device, and weatherproof enclosures are needed to protect it. Plus, a Raspberry Pi, either with built-in wireless connectivity or a WiFi dongle, is required for conducting computations.

##Software Subsystem

A practically usable weather forecast system is supposed to make reliable predictions for real-world multi-variable weather conditions. We apply Machine Learning techniques to suffice such generalization to unseen data. To this end, a high-quality dataset for training and evaluating the Machine Learning model is required, and a specially designed Machine Learning model would be developed on such a dataset. Once a well-trained system is obtained, we deploy the such model on portable devices with easy-to-use APIs.

#Criterion for Success

1. The weather measurement prototype with sensors should be able to accurately collect the temperature, humidity, and barometric pressure. etc.

2. A machine learning algorithm should be successfully trained to make predictions on the weather conditions: rainy, sunny, thunderstorm, etc.

3. Our system can forecast the weather in Haining, in real-time, and/or longer-period forecast.

4. The forecasted weather information could be demonstrated elegantly through some UI interface. A display screen would be a baseline, and an application on phones would be extra credit if time permitted.

5. Extra: Make our own weather dataset for Haining. If good, make it open-source.

#Work Distribution

**EE Student Zheyu Fu**:

-Design the sensor module circuit

-Development of visualization interface

**ECE Students Xuanyu Chen & Zhenting Qi**:

-Weather data collection and analysis

-Build and test Machine Learning model on Raspberry Pi

**ME Student Chenzhi Yuan**:

-Physical structure hardware design

-Proper distribution of the sensors to collect accurate data on temperature, humidity, barometric pressure, etc.