Model-Free Control Framework for Stability and Path Tracking of Autonomous Independent-Drive Vehicles

This article presents a model-free integrated control framework that uses deep reinforcement learning (DRL) to improve the stability and safety of four-wheel independently driven autonomous electric vehicles (4WID-AEVs). The proposed framework achieves precise path tracking and yaw motion control without relying on an accurate tire model. We introduce a novel hybrid DRL control strategy that effectively combines the Stanley controller with a DRL agent. This strategy enables trial-and-error learning through interaction with the vehicle environment, without requiring future state predictions or detailed mathematical models, ensuring adaptability, model independence, and superior real-time performance. Simulation results show that the strategy significantly improves lateral stability and tracking accuracy across various road conditions and speeds. Compared to the model predictive control (MPC), the model-free control method delivers better control performance and real-time responsiveness. Real-vehicle testing further validates the practical effectiveness of the proposed control strategy.