Robust Deep Reinforcement Learning-Based Coordinated Control of Direct Yaw Moment and Active Rear Wheel Steering for Stability of DDEVs

Distributed drive electric vehicles are characterized by their independently controllable multi-subsystems, enabling the realization of coordinated control amongst subsystems to improve the dynamic performance of a vehicle. This study aims to enhance the stability of vehicles by proposing a coordinated control strategy based on robust deep reinforcement learning for direct yaw moment control (DYC) and active rear wheel steering (ARS). First, an equivalent two-degrees-of-freedom vehicle dynamics model that considers the uncertainty of tire cornering stiffness is established. Then, a performance-guaranteed DYC and ARS controller is designed based on this model by using linear matrix inequality. Subsequently, considering the uncertainty of tire cornering stiffness, the maximum–minimum robust deep deterministic policy gradient algorithm is proposed to train the system. In this framework, the coordinated control strategy between the DYC and ARS controllers is formulated as a Markov decision process, with a reward function derived from the vehicle’s predictive state as well as the error between the actual and target states and stability region classifications. Finally, the proposed control strategy is validated through simulation and real vehicle experiments. Results demonstrate that the designed controller exhibits high robustness and effectiveness in enhancing the vehicle stability under various experimental conditions.