Reinforcement Learning-Based Motion Control of Four In-Wheel Motor-Actuated Electric Vehicles.

This article focuses on the development of a reinforcement learning (RL)-based adaptive multi-PID control strategy, termed TD3-QuadPID, for motion control of Four In-Wheel Motor Actuated Electric Vehicles (4IWMA EVs). The approach employs a model-free Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm to optimize four independent PID controllers—one per wheel—alongside steering control, enabling precise velocity and trajectory tracking while significantly improving energy efficiency without requiring explicit vehicle dynamic models. Simulation results in two driving scenarios—acceleration and braking on a rough slippery road, and a double lane-change maneuver—demonstrate that TD3-QuadPID outperforms a baseline PID controller in energy savings (up to 99.3%) and maintains comparable or improved tracking performance, with robustness validated in a separate test environment featuring unknown variations. The study highlights the potential of RL-based adaptive control for enhancing autonomous 4IWMA EV performance, while noting that full constraint satisfaction remains a challenge for future research.