Design of unmanned ground vehicle (UGV) path tracking controller based on reinforcement learning

This paper presents a unmanned ground vehicles (UGV) path tracking controller based on deep reinforcement learning (DRL), where a double deep Q-network (DDQN) algorithm is employed to train a deep neural network (DNN) for controlling the UGV to follow the desired path. The advantage of DDQN over deep Q-network (DQN) is that the DDQN uses two NNs, where one is working as a controller to generate actions for controlling the UGV, while the other is the target network to estimate the future rewards. The path tracking UGV kinematic is presented to determine the deviated distance and orientation between the UGV's pose and the desired path. White noise was added to the UGV wheels' speed for evaluating the robustness of the proposed controller. The simulation results illustrate that the trained controller enables the UGV to follow the desired trajectory in the presence of noisy actuation with high accuracy.