Improved deep reinforcement learning controls anti-lock braking via electro-hydraulic composite systems in in-wheel motor-driven electric vehicles

Abstract Anti-lock braking control for in-wheel motor-driven electric vehicles is challenging because electro-hydraulic composite braking combines fast and accurately controllable motor regenerative braking with slower hydraulic pressure regulation. Existing ABS strategies often disable regenerative braking during ABS intervention, which limits both braking performance and energy recovery. This paper proposes a delay-aware action-space-grouped Soft Actor-Critic controller, termed SAC-DA-ASG, for coordinated ABS control of the motor regenerative braking system and the hydraulic braking system. The proposed controller addresses two implementation issues. First, an action-space grouping scheme maps coordinated solenoid-valve operations into a single hydraulic mode, reducing invalid actuator combinations and the policy search space. Second, a delay-aware augmented state incorporates recent system states to compensate for heterogeneous actuator response delays and improve the Markov representation. The controller is trained under randomized road-adhesion and actuator conditions and is evaluated against baseline SAC variants and a Bosch-type ABS strategy. Simulation results show that SAC-DA-ASG keeps the wheel slip ratio closer to the target value of 0.15, reduces slip fluctuation and peak slip, and shortens braking distance under uniform and split- $$\mu$$ road conditions. Sensitivity tests under solenoid-valve delay and vehicle-mass perturbations further indicate robust closed-loop behavior. Real-time hardware-in-the-loop tests on a Speedgoat platform show close agreement with simulation in braking time, stopping distance, and slip-ratio regulation. These results demonstrate the feasibility of DRL-based coordinated electro-hydraulic ABS control for in-wheel motor-driven electric vehicles.