Shared Control for Autonomous Driving via Brain Emotional Learning Circuit Model: A Secure Reinforcement Learning Framework

The learning from intervention (LfI) approach has proven to be highly effective in enhancing the performance of reinforcement learning (RL) algorithms. It guides RL agents to adopt safe and efficient exploration strategies through real-time human intervention. However, in the context of autonomous driving, the quality of human intervention depends on the participants' driving skills, and there is a lack of theoretical guarantee in reality. To solve these problems, we propose a security perception reinforcement learning based on the Brain Emotional Learning Circuit Model (SPRL-BELCM). BELCM mimics human driving decisions, overcoming the limitations of human intervention. We adopt a safety assessment module, the Dynamic Potential Field (DPF), based on the artificial potential field (APF) theory. Using environmental information for real-time safety assessments, it achieves shared control through a dynamic control authority allocation mechanism.