Data-driven Adaptive MPC for Local Path Planning

To improve the obstacle avoidance safety and control accuracy of the model predictive control (MPC) for local path planning, a data-driven adaptive MPC local path planning method is proposed. Firstly, a data-driven MPC local path planning method is proposed to address the issue of reduced model accuracy of MPC algorithm under complex working conditions. In this method, the history data during the MPC path planning process of the unmanned vehicle is learned online utilizing sparse Gaussian process regression (SGPR), which improves the prediction accuracy of the model while ensuring the computational efficiency; Secondly, to overcome the limitation that the traditional obstacle avoidance constraint using single expansion coefficient in local path planning cannot balance the safe obstacle avoidance area and the path planning area, an adaptive expansion coefficient of obstacle avoidance constraint is proposed based on the relative position between the unmanned vehicle and the obstacle to achieve cautious obstacle avoidance. Finally, the path planning effect of the proposed method is verified through simulation and real vehicle experiment. The simulated results show that the proposed method reduces the maximum relative deviations between the predicted and actual longitudinal velocities, lateral velocities and yaw rates by 12, 82 and 63, respectively, compared with MPC. The average calculation time of the proposed method is 45.7 ms, and that of MPC is 37.9 ms. The vehicle test results show that the proposed method reduces the maximum relative deviations between the predicted and actual longitudinal velocities, lateral velocities and yaw rates by 4, 48 and 20, respectively, compared with MPC.