Adaptive model predictive controller for trajectory tracking and obstacle avoidance on autonomous vehicle

Abstract

Accurate vehicle trajectory tracking and collision free motion have become an active topic being discussed in autonomous vehicle research field. During an emergency obstacle avoidance manoeuvre condition, tyre force saturation greatly affects the trajectory tracking performance of the vehicle. Existing controllers such as conventional model predictive controller (MPC) and geometric controller (Stanley) need proper gain tuning to cope with this condition. This is due to the control gains were determined via linearization process at a certain targeted speed. Therefore, the control performance is limited considering the presence of speed variations with extreme manoeuvre trajectory. This paper proposes an adaptive MPC controller to solve aforementioned issues. First, optimized weighting gains for the steering control were obtained using PSO algorithm. The optimised weighting gains were then scheduled into the proposed adaptive MPC via a look-up table strategy. In this work, adaptive MPC was designed by using the linearization of the 7 degree-of-freedom (DOF) non-linear vehicle model. Here, the linearized model for controller design was updated based on the instantaneous longitudinal speed of the vehicle system plant. To evaluate adaptive MPC performance, comparisons with the adaptive Stanley controller and conventional MPC are conducted to analyse its effectiveness in low, middle and high-speed scenario. Simulation results showed that adaptive MPC improved the tracking error performance with respect to the speed variation in extreme collision avoidance manoeuvre. In high-speed manoeuvre (i.e. 25 m/s), lateral error improvement of 27.3% and 42.3% compared to conventional MPC controller and adaptive Stanley controller were obtained respectively.