Research on the Physics-Intelligence Hybrid Theory Based Dynamic Scenario Library Generation for Automated Vehicles.

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Systems Testing and Evaluation

Background:

• Automated vehicle development necessitates robust testing and evaluation systems for safety and performance.
• Current testing methods face challenges in efficiency and accuracy, particularly for complex driving scenarios.

Purpose of the Study:

• To propose a novel physics-intelligence hybrid theory-based dynamic scenario library generation method.
• To enhance the testing efficiency and accuracy for automated vehicles.
• To establish a general framework for dynamic scenario library generation.

Main Methods:

• Developed a parameterized scenario generation method utilizing dimension optimization to identify effective scenario elements.
• Constructed long-tail functions for specific Operational Design Domain (ODD) performance testing.
• Employed node optimization and sample expansion for low-dimensional scenario generation, and reinforcement learning for high-dimensional scenarios.

Main Results:

• The proposed method demonstrated superior efficiency and accuracy in generating dynamic scenarios for automated vehicle testing.
• Evaluated using naturalistic driving data (NDD) from intelligent electric vehicles, the method outperformed ideal testing libraries and raw NDD.
• Achieved significant improvements in both low- and high-dimensional scenario library generation.

Conclusions:

• The physics-intelligence hybrid approach offers a significant advancement in automated vehicle testing.
• This method provides a more efficient and accurate means of creating diverse and critical driving scenarios.
• The generated dynamic scenario libraries are crucial for the reliable deployment of autonomous driving technologies.