Learning-aided observer design for improving autonomous vehicle safety.

Area of Science:

• Automotive Engineering
• Control Systems
• Artificial Intelligence

Background:

• Automated vehicle safety and efficiency are critical during complex maneuvers.
• Existing stability index methods have limitations in detecting nuanced performance degradation.

Purpose of the Study:

• To introduce a novel observer design for enhancing autonomous vehicle safety and efficiency.
• To develop a method capable of detecting stability loss and performance degradation.

Main Methods:

• Utilized an observer design architecture evaluating lateral dynamic states.
• Integrated model-based and machine-learning-based (reinforcement learning) methods.
• Defined a lateral safety index based on an improvement vector.

Main Results:

• The novel observer provides efficient trajectory tracking and safety evaluation.
• The reinforcement learning-based observer achieves small observation error under nonlinear dynamics.
• The method simultaneously identifies safety and performance loss hazards.

Conclusions:

• The proposed observer-based safety evaluation method effectively detects stability and performance loss.
• This approach enhances the reliability and safety of autonomous vehicles during critical maneuvers.