Abstract
The strong coupling between lateral and longitudinal dynamics in autonomous vehicles presents a significant challenge for trajectory tracking control, especially under high-dynamic and complex conditions. To address this, this paper proposes a real-time optimal control method driven by a Reservoir Computing (RC)-based vehicle inverse dynamics model. The approach first involves training an RC network on a comprehensive vehicle dynamics dataset, covering multiple operating conditions, to learn the inverse mapping from accelerations to control commands. Second, an online correction mechanism incorporating Proportional-Derivative (PD) feedback is designed to dynamically adjust the desired acceleration inputs based on trajectory tracking errors. Finally, these corrected accelerations are fed into the trained RC network to rapidly compute high-precision control commands, completing the closed-loop tracking. Comprehensive simulations on double-lane-change, figure-eight, and Rössler chaotic trajectories demonstrate that the proposed method achieves high-precision tracking with remarkable computational efficiency and excellent robustness against control disturbances and sensor noise. Notably, moderate sensor noise exhibits trajectory-dependent performance enhancement, with system failure boundaries under combined disturbances clearly characterized.