A screenshot of our gazebo simulation shows a Neato preparing to dodge a blue ball moving in its direction.

Overview

Timeline: 5-week Final Project for Fall 2020 Computational Robotics Course at Olin College

Collaborators: Amy Phung, Nathan Faber, w. Professor Paul Ruvolo

The goal of this project was to train a neural network on a simulated Neato (mobile differential drive robot) to dodge balls. We generated data for training by driving the Neato and dodging balls ourselves, and used this data for supervised learning. We trained a standard feed forward network as well as a Long Short Term Memory (LSTM) network on this task, and tested our networks in simulation. See the project website for demos and more information.

My Contribution

My teammates and I designed our neural network sensor inputs and actuation outputs together. I helped to generate data for our networks, and built a 2D visualizer that provided insight as to how the network would react to different cases.