Autonomous Racing Robot: Hardware and Software Implementation

Abstract

Artificial intelligence and machine learning are used to solve more and more different and complex problems. One popular machine learning subfield is imitation learning, where robot learns to solve a specified task using demonstrations from a human expert. Imitation learning is popular because it is easy to show an example solution to a task and let the machine figure out how to replicate it. One of the most used methods of imitation learning is behavioural cloning, where robot learns from expert demonstrations to map observations to actions. This thesis proposes a simple hardware and software implementation for a small-scale autonomous robot. Behavioural cloning is used to teach a robot to autonomously drive around a small test track. Training data contains about 50 minutes of video, which corresponds to about 58000 images as well as control commands for those images. Training image set consists of 3 different classes, which correspond to the three different movement commands. The tests show that the proposed implementation works well on the chosen test track.