Hybrid Supervised and Reinforcement Learning for Motion-Sickness-Aware Path Tracking in Autonomous Vehicles

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Path tracking is critical for autonomous driving (AD) safety, comfort, and efficiency.
• Existing methods struggle to balance tracking accuracy with computational cost and do not address motion sickness (MS).
• MS is a growing concern in AD as passengers engage in non-driving activities.

Purpose of the Study:

• To develop a novel framework, Hybrid Supervised-Reinforcement Learning (HSRL), for AD path tracking.
• To reduce passenger discomfort and motion sickness (MS) while maintaining high tracking accuracy and computational efficiency.
• To address the limitations of current path-tracking controllers.

Main Methods:

• Implemented a Hybrid Supervised-Reinforcement Learning (HSRL) framework.
• Utilized expert data-guided supervised learning to optimize the path-tracking model, addressing Reinforcement Learning (RL) sample efficiency.
• Integrated a passenger MS mechanism into the RL's multi-objective reward function for enhanced robustness and comfort.

Main Results:

• The HSRL framework demonstrated superior performance compared to Proportional-Integral-Derivative (PID) and Model Predictive Control (MPC).
• Achieved high-precision path tracking with improved computational efficiency.
• Significantly reduced the cumulative Motion Sickness Dose Value (MSDV) for passengers across various scenarios.

Conclusions:

• HSRL effectively balances high-precision path tracking with passenger comfort optimization in autonomous driving.
• The proposed framework offers a computationally efficient solution for AD path tracking.
• HSRL presents a promising approach for mitigating motion sickness in self-driving vehicles.