Trajectory tracking for 4WIS vehicles under high-curvature conditions using model-based proximal policy optimization

The 4-wheel independent steering (4WIS) vehicles are capable of highly flexible maneuvers, but trajectory tracking under high-curvature conditions remains challenging due to tire nonlinear characteristics and dynamic coupling. This paper proposes a novel model-based proximal policy optimization (MBPPO) framework that integrates an accurate nonlinear dynamic model with reinforcement learning. An accurate vehicle dynamics model is established, where the relationship between the steering instantaneous center offset and the tire nonlinear characteristics is analyzed to define a curvature error combining steering angles and slip angles. Building on this formulation, the output of a model predictive control (MPC) trajectory tracking controller is embedded into the proximal policy optimization (PPO) algorithm as ‘expert demonstrations’. This integration enhances sampling efficiency, accelerates convergence, and improves robustness compared with conventional PPO and MPC approaches. Co-simulation in CarSim/Simulink and Hardware-In-Loop (HIL) tests verify that the proposed MBPPO controller achieves faster learning, reduced lateral deviation, and smaller heading angle errors under high-curvature conditions.