Driving Decision-Making Method for Autonomous Vehicles Based on Anterior Cingulate Cortex Neural Regulation Mechanism

Reinforcement learning algorithms have been widely applied in the field of autonomous driving. However, the choice of policy during the reinforcement learning process significantly impacts learning outcomes. Existing stochastic policy methods lack flexibility, preventing autonomous vehicles from quickly learning optimal strategies in complex traffic environments. This paper proposes a model that simulates the regulatory mechanism of the Anterior Cingulate Cortex (ACC) to balance exploration and exploitation. The model observes through the Medial Prefrontal Cortex (MPFC) module, evaluates using graph convolutional neural networks, and transmits dopamine signals of reward errors to the ACC module. Feedback neurons compute vigilance, and the LPFC adjusts action probabilities based on vigilance and action values. Experimental results validate that this approach achieves a balance in action selection strategies for autonomous vehicles, thereby improving the quality of actions taken by the agent.