-
ICSE 2026 submission · 2026 · CCF-A, co-first author
ADS Bug Localization Study
Studies functional-code-level ADS bug localization for connecting observed driving failures back to likely implementation causes.
-
TOSEM under review · 2026 · CCF-A, first author
ADS Test-Scenario Coverage Analysis
Defines multidimensional scenario coverage and coverage-guided multi-agent generation; in million-scale tests, it finds up to 556% more unique failure patterns than baselines and reduces downstream accident rates by 75-78% after safety hardening.
-
FSE 2026 Tool Demonstrations · 2026 · CCF-A, student first author, Tool Demonstration
Xiaodong Zhang, Songyang Yan, Ming Fan, Zijiang Yang
CapCo builds an Apollo-CARLA bridge for closed-loop co-simulation and scenario fuzzing, reducing manual configuration and supporting batch testing, result analysis, and reproducible tool-demonstration workflows.
DOI · researchr
-
FM 2026 / Formal Methods, LNCS 16557 · 2026 · CCF-A, co-first author, oral presentation
Xiaodong Zhang, Songyang Yan, Zijiang Yang
ConFixer uses STL robustness signals to localize and repair Apollo configuration bugs, fixing 173 failed scenarios on Apollo 7.0 without introducing regressions across 1,680 passing scenarios.
DOI · Springer
-
2025 IEEE International Conference on Robotics and Automation (ICRA) · 2025 · CCF-B
Haojie Xin, Xiaodong Zhang, Songyang Yan, Jun Sun, Zijiang Yang
The paper combines structural-causal counterfactual augmentation, contrastive learning, and shortcut elimination to reduce causal confusion in imitation-based planning, reaching 92.72 CLS-NR / 91.38 CLS-R on nuPlan val14 and improving 6 of 8 long-tail scenario types.
DOI · DBLP
-
Proceedings of the ACM on Software Engineering, 2(FSE) · 2025 · CCF-A, first author, oral presentation
Songyang Yan, Xiaodong Zhang, Kunkun Hao, Haojie Xin, Yonggang Luo, Jucheng Yang, Ming Fan, Chao Yang, Jun Sun, Zijiang Yang
OSG generates automated driving system test suites with controllable risk levels while preserving naturalness and diversity; on Apollo, it raises the accident rate by 92.97% over SOTA and speeds analysis by about 4x.
DOI · FSE 2025 · DBLP
-
IEEE Transactions on Intelligent Vehicles, 9(9) · 2024 · JCR Q1, IF 14.3
Kunkun Hao, Wen Cui, Yonggang Luo, Lecheng Xie, Yuqiao Bai, Jucheng Yang, Songyang Yan, Yuxi Pan, Zijiang Yang
The paper uses naturalistic human-driving priors to constrain PPO-based adversarial generation, producing more realistic safety-critical scenarios and improving generation efficiency by 44% over baselines.
DOI · DBLP
-
2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC) · 2024
Haojie Xin, Xiaodong Zhang, Renzhi Tang, Songyang Yan, Qianrui Zhao, Chunze Yang, Wen Cui, Zijiang Yang
LitSim preserves log-replay realism while resolving conflicts for long-horizon interactive traffic simulation.
DOI · arXiv
-
ACCES 2022 / Lecture Notes in Electrical Engineering 1014 · 2023
Songyang Yan, Yuxi Pan, Dan Shan, Changle Lin
SDG provides a software-in-the-loop simulation platform with scenario description and driving-behavior verification for automated driving system development.
DOI · Springer
-
Proceedings of the 15th Workshop on Search-Based Software Testing · 2022
Songyang Yan, Ming Fan
AdaFrenetic was an SBST 2022 tool-competition entry that reduces invalid automated driving test cases by adapting road points.
DOI · DBLP
