Safety Decision-Making for Autonomous Vehicles Integrating Passenger Physiological States by fNIRS.

In recent years, several serious traffic accidents have exposed the severity of safety issues in autonomous driving technology. Traditional decision-making methods are unable to address potential risky behaviors caused by the functional insufficiencies or machine performance limitations, and human intervention is still needed. This study proposes an intelligent safety decision-making algorithm with passengers' risk assessment by analyzing passenger physiological states online using functional near-infrared spectroscopy (fNIRS). This algorithm is developed based on twin-delayed deep deterministic policy gradient (TD3), and it can overcome the functional insufficiencies of traditional TD3 and guide TD3 using passengers' risk assessment by analyzing passenger physiological states online while confronting risky scenarios. Three experiments have been conducted in autonomous emergency braking, front vehicle cutting-in, and pedestrian crossing scenarios. The results show that the proposed algorithm demonstrates faster convergence and superior safety and comfort performance compared with traditional TD3. This study highlights the applicability of fNIRS technology in enhancing the safety and comfort of autonomous vehicles in the future.