Tabular Q-learning Based Reinforcement Learning Agent for Autonomous Vehicle Drift Initiation and Stabilization

Abstract

This paper aims to report on novel research results about developing a reinforcement learning agent for steady-state vehicle drift motion control. Based on the previous results of this research, the primary goal was to eliminate the problems causing learning instability experienced with the Soft Actor-Critic (SAC) algorithm applying Tabular Q-learning in this work. Trained in a MATLAB/Simulink-based simulation environment, the resulting agent succeeded in this task while being able to smoothly operate the vehicle to achieve and retain the desired target drift state, regardless of the discreet nature of the algorithm used for solving an inherently continuous task.