【摘要】 针对全线控分布式驱动车辆（Robo-Car）在复杂运营场景下存在的侧倾补偿能力不足与实时决策矛盾问题，本文提出一种基于强化学习的横摆-侧倾力矩协同决策与转矩解耦控制方法（MLC-RL）。该方法旨在结合强化学习的端到端决策优势，消除传统控制对显式模型的依赖，有效提升计算效率，并利用轮毂电机反力矩被动抗侧倾机制，补偿该车型无主动悬架的机械局限，从而实现对车辆横摆-侧倾稳定性的实时协同控制。MLC-RL采用分层架构：上层基于强化学习策略网络，直接从CarSim车辆动态数据中学习附加横摆力矩与抗侧倾力矩的协同决策；下层通过转矩解耦控制实现四轮转矩分配。MATLAB/Simulink-CarSim联合仿真结果表明，该方法在有效保证Robo-Car横摆稳定性的同时，显著提升了侧倾抑制效果与乘坐舒适性，为全线控分布式驱动车辆的稳定性控制提供了有效解决方案。更多还原

【Abstract】 For the insufficiency of roll compensation and the contradiction of real-time decision-making in complex operational scenarios for all-wheel-controlled distributed drive vehicles(Robo-Car), this paper proposes a yaw-roll moment coordinated decision-making and torque decoupling control method based on reinforcement learning(MLC-RL). This method aims to combine the end-to-end decision-making advantage of reinforcement learning, eliminate the dependence of traditional control on explicit models, effectively improve computational efficiency, and utilize the passive anti-roll mechanism of hub motor reaction torque to compensate for the mechanical limitation of this vehicle type without active suspension, thereby achieving real-time coordinated control of vehicle yaw-roll stability. MLC-RL adopts a hierarchical architecture. The upper layer is based on a reinforcement learning policy network, directly learning the coordinated decision-making of additional yaw moment and anti-roll moment from CarSim vehicle dynamic data. The lower layer realizes four-wheel torque distribution through torque decoupling control. The MATLAB/Simulink-CarSim co-simulation results show that this method effectively guarantees the yaw stability of Robo-Car while significantly improving the roll suppression performance and ride comfort, providing an effective solution for the stability control of all-wheel-controlled distributed drive vehicles.更多还原