Modeling of a DeepRacer

Abstract

DeepRacer is 1/18th model race car developed by Amazon. It clouds control by an Artificial Intelligence (AI) model. It training and simulation are conduct on Amazon Web Service (AWS) cloud platform with it dedicated AWS DeepRacer Console. With of Machine Learning (ML) method, Reinforcement Learning (RL). Eventually, the model cloud be converted from simulation to real (S2R) and the DeepRacer cloud race on a track. The project explored the principle of RL on AWS and key information of DeepRacer. The aim of this project is to train model that cloud drive DeepRacer as fast as possible. Three approach of training method have been conducted in this project. Approach I: Center Line and Minimal Speed, Approach II: Optimal Racing and Approach III: Simplify Optimal Racing. We will compare these three approach in terms of the methodology, result and performance. Moreover, a S2R experiment had also been conducted.