[论文解读] Coverage-Guided Road Selection and Prioritization for Efficient Testing in Autonomous Driving Systems

Qurban Ali, Andrea Stocco+2 more|arXiv (Cornell University)|Jan 13, 2026

Autonomous Vehicle Technology and Safety被引用 0

一句话总结

一种基于行为的测试优先级框架通过几何与驾驶动态将路段聚类，从而在减少冗余测试的同时保留故障暴露， enabling earlier failure detection.

ABSTRACT

Autonomous Driving Assistance Systems (ADAS) rely on extensive testing to ensure safety and reliability, yet road scenario datasets often contain redundant cases that slow down the testing process without improving fault detection. To address this issue, we present a novel test prioritization framework that reduces redundancy while preserving geometric and behavioral diversity. Road scenarios are clustered based on geometric and dynamic features of the ADAS driving behavior, from which representative cases are selected to guarantee coverage. Roads are finally prioritized based on geometric complexity, driving difficulty, and historical failures, ensuring that the most critical and challenging tests are executed first. We evaluate our framework on the OPENCAT dataset and the Udacity self-driving car simulator using two ADAS models. On average, our approach achieves an 89% reduction in test suite size while retaining an average of 79% of failed road scenarios. The prioritization strategy improves early failure detection by up to 95x compared to random baselines.

研究动机与目标

• Mitigate redundancy in road-based ADAS test datasets without sacrificing fault coverage.
• Develop a behavior-aware framework that segments roads by geometry and dynamics for clustering.
• Prioritize and sequence test cases to reveal failures earlier while preserving diversity and coverage.

提出的方法

• Curvature-based road segmentation combining straight, left- and right-curve classifications.
• Geometric matching with Dynamic Time Warping to filter redundant sections and identify inclusion relationships.
• Hybrid distance metric combining geometric and dynamic features for agglomerative clustering.
• Coverage-based selection to choose representative sections covering all clusters.
• Multi-criteria test prioritization using geometric, dynamic, and historical failure signals to order tests.

实验结果

研究问题

• RQ1 RQ1: How effective is the coverage-based selection in reducing tests while retaining failures?
• RQ2 RQ2: Does incorporating dynamic driving behavior improve test selection and fault detection compared with geometry-only methods?
• RQ3 RQ3: How effective is the proposed prioritization in identifying critical scenarios earlier than random ordering?
• RQ4 RQ4: Do tests selected for one ADAS model transfer failure revealing power to architecturally different models?

主要发现

Campaign	Total No. Tests	No. Failed	Selected Tests	Reduction %	FRR Selected%	EFD	EFD10	APFD
Ambiegen Campaign 2	973	11	147	85%	45%	0.17%	45%	1.04%	0.92
Ambiegen Campaign 3	964	9	206	79%	89%	0.20%	80%	1.04%	0.95
Ambiegen Campaign 4	965	5	178	82%	80%	0.10%	80%	1.04%	0.93
Ambiegen Campaign 5	958	10	167	83%	80%	0.18%	70%	1.04%	0.91
Ambiegen Campaign 6	959	9	179	81%	78%	0.18%	70%	1.04%	0.89
Ambiegen Campaign 7	963	10	197	80%	70%	0.21%	60%	1.04%	0.96
Ambiegen Campaign 8	952	11	176	82%	91%	0.21%	91%	1.05%	0.92
Ambiegen Campaign 9	953	4	187	80%	100%	0.08%	75%	1.05%	0.97
Ambiegen Campaign 10	971	18	?	?	?	?	?	?	?
OpenCat Campaigns (aggregate)	323?	?	?	?	?	?	?	?	?

• Test suite reductions up to 89% were achieved while retaining an average of 79% of failed road scenarios.
• Prioritization improved early failure detection by up to 95× (median 78×) compared to random baselines.
• APFD scores were high (0.9 vs 0.2 for random ordering), indicating strong fault-detection efficiency.
• Dynamic data integration with geometry improved test selection effectiveness over geometry-only approaches.
• Representative clustering preserved both geometric diversity and driving behavior complexity.