A Bayesian Driver Agent Model for Autonomous Vehicles System Based on Knowledge-Aware and Real-Time Data.

Area of Science:

• Autonomous Driving Systems
• Artificial Intelligence
• Cognitive Psychology

Background:

• Modeling driver decision-making is crucial for autonomous vehicle safety and efficiency.
• Uncertainty in urban traffic trajectories challenges current autonomous vehicle systems.
• Existing methods like end-to-end learning and rule-based systems have limitations.

Purpose of the Study:

• To propose a novel Bayesian driver agent (BDA) model for autonomous vehicles.
• To develop a vision-based system inspired by human cognitive psychology.
• To improve the accuracy and robustness of autonomous vehicle decision-making.

Main Methods:

• A vision-based approach separating scene recognition and decision inference modules.
• Utilizing a Convolutional Neural Network (CNN) for multi-task learning in perception.
• Employing a dynamic Bayesian network (DBN) for decision inference based on scene features.

Main Results:

• The BDA model effectively extracts traffic scene features.
• Accurate prediction of human driver decision-making probability distributions.
• Achieved a high intraclass correlation coefficient (ICC) of 0.984 in lane-changing scenarios.

Conclusions:

• The proposed BDA model offers a robust framework for autonomous driving decision-making.
• This approach integrates perception and inference inspired by human cognition.
• Demonstrates significant potential for enhancing autonomous vehicle systems.