Beach Cleaning Robot

Undergraduate Major Qualifying Project

Contributions:

Identified, designed, and evaluated methods of conveyor-sifting, and sensor-equipped debris storage and expulsion system
Trained machine learning-based object recognition and redesigned electrical layout
Consulted on hight-adjustable trash collection mechanism

Project Overview

This multi-year group project (members change each year) aimed to help clean beaches without human direction or causing further ecological damage. There are currently beach-cleaning robots in use that sift through sand to collect debris in a similar way to ours, but none (at the time) were autonomous. Our robot system utilizes computer vision to autonomously identify and collect trash while avoiding organic debris, detects when its full, traverses uneven terrain and can adjust the height of its collection mechanism.

Asymmetric Sifting.MOV

Sifter Assembly

The sifter uses a 12V motor with an rpm of 150
Miniature drawer slides allow the sifter to translate on a stationary frame
Chicken wire stapled to plywood waveform is used as sifting mesh

Asymmetrically Oscillating Sifter

Utilizing a four-bar crank-slider mechanism, asymmetric oscillation was induced to convey objects too large to fit within sifter troths. This SolidWorks simulation shows the required torque from the motor. An analysis on the sifter motion was conducted with Norton Linkages.

Required Torque Per Time (bottom left)

Velocity and Speed Per Time

Early Catch Pan Concept FBD

Required Torque Calculation Guide

Sensory Catch Pan

While the grandfathered architecture didn’t permit much undercarriage clearance, a small catch pan that detects how much debris has been collected based on the electrical impedance of the dumping mechanism motor was added below the robot’s sifter and battery mount.

Computer Vision Machine Learning Model

The current iteration of the computer vision model is an AutoML Edge model created and trained using Google’s Vertex AI. It has been trained on an image data set of 196 items. These images fall into one of nine categories: Sand, seaweed, shells, starfish, bottles, cans, cigarettes, plastic bags, or wrappers. The precision (percentage of correct predictions) is 98.3%. The recall (percentage of relevant data points identified) is 84.1%.

Confusion Matrix

System Circuit Diagram

After numerous malfunctions and short circuits, I was put in charge of rewiring our electrical systems and documenting changes using Fritzing, an electrical diagram software chosen by the previous year’s team.

Google Sites

Report abuse