Using Least-Square Policy Iteration to Online Optimize the Parameters of ALV´s Speed Controller

Due to the highly non-linear properties of the longitudinal dynamics of autonomous land vehicles (ALVs), it is difficult to tune the parameters of a speed controller for the autonomous driving of ALVs. Aiming at this problem, in this paper, a novel learning-Based speed controller is proposed, which is composed of a time-varying proportional-integral (PI) control structure and a learning-Based learning module. A near-optimal policy is obtained by least-square policy iteration (LSPI), which is an approximate policy iteration method. The learning-Based module uses the near-optimal policy to realize online tuning of the PI coefficients. The simulation results show that the proposed controller can optimize the control performance by combining different non-optimal coefficients of the PI structure.