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

Area of Science:

• Robotics and Artificial Intelligence
• Biomedical Engineering
• Transportation Safety

Background:

• Autonomous driving technology faces safety challenges due to functional limitations, necessitating human intervention.
• Traditional decision-making algorithms struggle with unpredictable risky behaviors in autonomous vehicles.
• Passenger well-being and risk perception are critical factors for safe autonomous driving.

Purpose of the Study:

• To develop an intelligent safety decision-making algorithm for autonomous vehicles that incorporates passenger risk assessment.
• To enhance the safety and comfort of autonomous driving by analyzing passenger physiological states in real-time.
• To overcome the limitations of traditional decision-making methods in autonomous driving systems.

Main Methods:

• Proposed an intelligent safety decision-making algorithm based on twin-delayed deep deterministic policy gradient (TD3).
• Integrated passenger risk assessment by analyzing physiological states online using functional near-infrared spectroscopy (fNIRS).
• Conducted experiments in autonomous emergency braking, front vehicle cutting-in, and pedestrian crossing scenarios.

Main Results:

• The proposed algorithm demonstrated faster convergence compared to traditional TD3.
• The algorithm achieved superior safety performance in critical driving scenarios.
• Enhanced comfort levels were observed with the new decision-making approach.

Conclusions:

• The intelligent safety algorithm effectively enhances autonomous vehicle decision-making by incorporating passenger risk assessment.
• Functional near-infrared spectroscopy (fNIRS) is a viable technology for real-time passenger monitoring in autonomous vehicles.
• This approach significantly improves both the safety and comfort of autonomous driving systems.