The Path-Tracking Method Based on Deep Deterministic Policy Gradient and Applied for High-Speed Driving

This paper proposes a high-speed path-tracking algorithm based on the deep deterministic policy gradient (DDPG) strategy with curvature adaptive ability applied to automatic driving. The tracking controller contains two mechanisms: the pure pursuit (PP) module and the model predictive control (MPC) module. The PP part works well in the small curvature road conditions, while the MPC can solve more complex problems. Thus, DDPG triggers the decision of which mechanism to use: PP works on regular roads or scenes with relatively small curvature, while it is switched to MPC control for complex scenarios by building the testing environments in CarMaker to validate the control performance of our method.