Intervention Feasible Region and Driver Risk Capacity Aware Human-Machine Collaborative Safe Trajectory Planning

Area of Science:

• Robotics and Autonomous Systems
• Artificial Intelligence
• Human-Machine Interaction

Background:

• Inconsistent tracking goals between humans and machines are a major cause of accidents in collaborative driving.
• Ensuring driving safety requires addressing both objective physical constraints and subjective driver preferences.

Purpose of the Study:

• To propose a human-machine collaborative safe trajectory planning algorithm that enhances driving safety.
• To develop methods for assessing intervention feasible region (IFR) and driver risk capacity (DRC) for personalized safety.

Main Methods:

• Constructed IFR solution using CTRA model predictions and boundary constraints.
• Developed a bidirectional subjective perceived risk quantification for personalized DRC assessment.
• Designed a risk-adaptive safe deep reinforcement learning (DRL) algorithm (RAS-β-SAC) incorporating beta distribution and risk relabeling.

Main Results:

• The proposed algorithm ensures planned trajectories meet safety constraints and satisfy personalized DRC.
• Achieved dual safety guarantees: objective physical safety and subjective driver acceptability.
• Reduced human-machine conflicts through risk-adaptive planning and personalized trajectory adjustments.

Conclusions:

• The developed trajectory planning algorithm effectively enhances safety in human-machine collaborative driving.
• Integrating IFR and DRC provides a robust framework for safe and acceptable autonomous navigation.
• RAS-β-SAC demonstrates the potential of DRL for personalized and safe collaborative driving systems.