Balancing Tracking Accuracy and Stability in Distributed-Drive Off-Road Autonomous Vehicles: An Adaptive TD3 Framework

Off-road vehicles are designed for operation on unstructured terrain under dynamic environmental conditions. The combination of unpredictable surface variations and complex topographical features presents significant challenges to maintaining vehicular stability during off-road navigation. To address these challenges, a reinforcement learning-based Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm is proposed in this study to ensure accurate and stable path tracking for off-road vehicles operating in complex terrain conditions. A distributed-drive off-road vehicle dynamics model is developed through systematic analysis of wheel-terrain interaction characteristics, which not only facilitates the creation of a unified pressure-shear model with wheel subsidence dynamics but also establishes dual evaluation index for roll and sideslip stability. The Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm is formulated through reinforcement learning, wherein the state space, action space, reward function, and network architecture are systematically defined. Additionally, tracking stability is effectively attained via integrated control of active steering and an additional yaw moment. The effectiveness of the proposed control strategy is verified through simulation and hardware-in-the-loop testing, and is compared with traditional MPC control.