Joint acceleration and steering control via deep Q-network for high-precision path tracking of autonomous vehicles

This study presents a deep reinforcement learning framework for high-precision autonomous vehicle path tracking under stochastic driving environments. The path-tracking problem is formulated as a task-oriented Markov decision process (MDP) with a discretized dual-action structure that jointly models acceleration and steering decisions, enabling coordinated optimization of longitudinal and lateral vehicle dynamics within an unified control framework. The proposed joint deep Q-network (J-DQN) learns the control policy directly through interaction with the environment without relying on explicit system models. Experimental evaluations demonstrate that the proposed approach achieves superior tracking accuracy, stability, and robustness compared with representative baseline methods under realistic driving scenarios.