Reinforcement Learning-Enhanced Motion Control for Distributed-Drive Intelligent Vehicles With Performance Variation of Driving Units

This article investigates the reinforcement learning (RL)-enhanced motion control problem for distributed-drive intelligent vehicles facing performance variations in their driving units. To tackle this challenge, we first analyze the sources of performance variations in driving units and model these variations as unknown degradation coefficients within the vehicle dynamics framework. In addition, we integrate the deep deterministic policy gradient algorithm with a well-established control allocation method into a hierarchical control architecture to mitigate the effects of driving unit performance fluctuations. By combining (RL-based control with preexisting control strategies, our framework effectively balances vehicle lateral stability and adaptability to driving unit performance variations. To validate the proposed approach, we conduct both numerical simulations and hardware-in-the-loop (HiL) experiments. The results demonstrate that the RL-based controller can adaptively handle diverse performance degradation scenarios and significantly enhance tracking accuracy. In addition, HiL experiments confirm the real-time implementability of the RL algorithm, achieving satisfactory control performance under practical operating conditions.