Deep Reinforcement Learning for Autonomous Dynamic Skid Steer Vehicle Trajectory Tracking

Designing controllers for skid-steered wheeled robots is complex due to the interaction of tires with ground and wheel slip skid-steer driving mechanism, leading nonlinear dynamics. Due recent success reinforcement learning algorithms mobile robot control, Deep Deterministic Policy Gradients (DDPG) was successfully implemented an algorithm designed continuous control problems. The dynamics vehicle model were dealt advantages deep neural networks leveraged their generalizability. Reinforcement used gather information train agent in unsupervised manner. performance trained policy on six degrees freedom dynamic simulation demonstrated force interactions. system met requirement stay within distance half width from reference paths.