An online reinforcement learning-based conflict resolution approach for human-vehicle cooperative systems

Conflicts between driver and automation remain a key challenge in cooperative driving systems, as prolonged differences in intentions can degrade human experience, reduce trust, and compromise driving safety. In this paper, a conflict resolution approach is proposed for human-vehicle cooperative systems to minimize errors in trajectory tracking and conflicts between driver and automation. By considering driver-in-the-loop, a vehicle-driver-conflict model is introduced, and then an online reinforcement learning algorithm is developed to calculate the automated steering angle for cooperative systems, which can assist humans even in conflict situations. Based on Lyapunov's theory, the states of the vehicle-driver-conflict model are guaranteed to be uniformly ultimately bounded and the convergence of the control signal and cost function to their optimal values is verified. Finally, the effectiveness of the proposed method is demonstrated and compared with other methods through numerical simulations.