ScenePilot: Controllable Boundary-Driven Critical Scenario Generation for Autonomous Driving
ScenePilot is a new framework designed for generating critical scenarios in autonomous driving. It focuses on creating boundary-driven scenarios that are physically solvable yet challenging for autonomous systems. The approach combines physical feasibility with risk prediction to enhance the testing of autonomous vehicles.
- ▪ScenePilot targets scenarios that are physically solvable but still cause failures in autonomous driving systems.
- ▪The framework uses constrained multi-objective reinforcement learning to generate critical scenarios.
- ▪Experiments show that ScenePilot increases collision rates while maintaining physical validity.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.21168 (cs) [Submitted on 20 May 2026] Title:ScenePilot: Controllable Boundary-Driven Critical Scenario Generation for Autonomous Driving Authors:Qiyu Ruan, Yuxuan Wang, He Li, Zhenning Li, Cheng-zhong Xu View a PDF of the paper titled ScenePilot: Controllable Boundary-Driven Critical Scenario Generation for Autonomous Driving, by Qiyu Ruan and 4 other authors View PDF HTML (experimental) Abstract:Safety-critical scenarios are central to evaluating autonomous driving systems, yet their rarity in naturalistic logs makes simulation-based stress testing indispensable.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
