Modeling of a DeepRacer

Abstract

Amazon Web Services DeepRacer is used to perform self-driving races by different control algorithms based on deep learning. It is a 1/18 scale racecar produced by AWS, and developers can use it to learn the Reinforcement Learning Model in a very interesting way. Developers can also quickly develop self-driving car machine learning models then test their control algorithm model on the AWS cloud platform with virtual DeepRacer and racetracks in a 3D simulator. In this project, a model that represents the motion of DeepRacer is obtained and it was used to design 2 control systems by using the theories of PID control and MPC. The control systems were improved and redesigned after the simulation was performed many times. Afterwards, the perfect performance of path tracking was achieved. The method to improve the latency problem when using PID controllers was mentioned in this paper. The simulation environment was established on Python and the result was discussed and analyzed at the end of this paper.