Make full use of testing information: An integrated accelerated testing and evaluation method for autonomous driving...

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Software Testing

Background:

• Autonomous driving systems (ADSs) require rigorous testing and evaluation to ensure the safety of the intended functionality (SOTIF).
• Current testing methods generate abundant data but often fail to fully leverage intermediate information for comprehensive evaluation.
• Identifying hazardous operational domains is crucial for validating ADSs.

Purpose of the Study:

• To propose an Integrated accelerated Testing and Evaluation Method (ITEM) for ADSs.
• To enhance the accuracy of hazardous domain identification by utilizing intermediate testing information.
• To improve the efficiency and effectiveness of SOTIF assessment for ADSs.

Main Methods:

• Development of the Integrated accelerated Testing and Evaluation Method (ITEM) based on Monte Carlo Tree Search (MCTS) and a dual surrogates testing framework.
• Integration of intermediate testing information, including tree structure and subspace relationships, into the evaluation stage.
• Improvement of the Upper Confidence Bound (UCB) calculation for focused exploration of hazardous domain boundaries and a novel stopping condition based on search algorithm convergence.

Main Results:

• ITEM successfully identified hazardous domains in both low- and high-dimensional scenarios, irrespective of domain shape.
• The method demonstrated generality and potential for robust safety evaluation of ADSs.
• Ablation and comparative experiments confirmed the effectiveness of the proposed improvements and the superiority of ITEM.

Conclusions:

• The proposed ITEM effectively utilizes intermediate testing data for accurate hazardous domain identification in ADSs.
• ITEM offers a generalizable and powerful approach for SOTIF assessment, crucial for the safe deployment of autonomous driving.
• The enhanced UCB calculation and convergence-based stopping condition contribute to focused and efficient safety evaluations.