Trash accumulation around bins is a common issue due to improper disposal habits. In this project, to help alleviate this issue in indoor settings, we developed the litter-rescuing robot (LR-Bot), an autonomous litter-picking robot that leverages the Jetson Nano 2GB for edge computing and a 4 DoF robotic arm with a suction gripper for trash picking. To evaluate our robot, we collected a custom dataset named TrashDet, featuring images from 9 common indoor trash categories, each annotated with bounding boxes. After testing three popular detection algorithms on Jetson Nano, we selected YOLOv5s as our object detector for its low latency and high accuracy. The robot uses a front-mounted camera to identify and position target trash, achieving an inference speed of 12 FPS, complemented by an ultrasonic sensor for distance measurements. It is controlled through a customized task-specific control loop. In our testing environments, our LR-Bot achieved an overall success rate of 78.33% in complete litter rescue tasks across six trash categories over 120 trials.

Demo

For detailed description, you can see in the link below.

Teammate: Yuan Qing