Research map / Probabilistic and scientific modeling / Molecules and proteins
CCFM: Collision-Constrained Flow Matching for Safety-Critical Scenario Generation
Ke Li, Kaidi Liang, Yuxin Ding, Debojyoti Biswas, Xianbiao Hu, Ruwen Qin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace soft collision guidance with Gauss-Newton manifold projection (ablation) | CCFM · nuScenes closed-loop | collision rate (%): 10.2 -> 46.4 | Improved |
| add soft collision guidance J_col to flow matching (ablation) | FM · nuScenes closed-loop | collision rate (%): 8.2 -> 10.2 | No clear effect |
| add soft guidance J_col to flow matching (ablation) | FM · nuScenes closed-loop | off-road rate (%): 2.3 -> 6.2 | Hurt |
| add GN projection to flow matching (ablation) | FM · nuScenes closed-loop | realism (RM): 0.77 -> 0.78 | No clear effect |
| use Heuristic Collision Selector instead of fixed single type (ablation) | CCFM · nuScenes closed-loop | type match (%): 70.7 -> 84.3 | Improved |
| use HCS instead of cut-in only selection (ablation) | CCFM · nuScenes closed-loop | collision rate (%): 40.0 -> 46.4 | Improved |
| add relative heading constraint to contact constraint (ablation) | CCFM · nuScenes closed-loop | type match (%) / collision rate: TM 82.1 -> 83.8; CR 51.6 -> 48.4 | Mixed |
| add severity constraint (ablation) | CCFM · nuScenes closed-loop | relative speed (m/s): 2.6 -> 2.8 | Improved |
| use CCFM instead of SAFE-SIM | CCFM · nuScenes closed-loop (horizon 80) | collision rate (%): 25.8 -> 46.4 | Improved |
| use CCFM instead of SAFE-SIM | CCFM · nuScenes closed-loop (horizon 80) | realism (RM): 0.63 -> 0.78 | Hurt |
| switch ego planner from lane-graph to PDM | CCFM · nuScenes closed-loop | collision rate (%): 46.4 -> 36.1 | Hurt |
| extend rollout horizon from 80 to 200 frames | CCFM · nuScenes closed-loop | collision rate (%): 46.4 -> 59.9 | Improved |
Closest papers by meaning
- ScenePilot: Controllable Boundary-Driven Critical Scenario Generation for Autonomous Driving (ICML 2026 · similarity 0.89)
- GuideFlow: Constraint-Guided Flow Matching for Planning in End-to-End Autonomous Driving (CVPR 2026 · similarity 0.87)
- DiffScene: Diffusion-Based Safety-Critical Scenario Generation for Autonomous Vehicles (AAAI 2025 · similarity 0.86)
- NAVSIM: Data-Driven Non-Reactive Autonomous Vehicle Simulation and Benchmarking (NeurIPS 2024 · similarity 0.85)
- ChatScene: Knowledge-Enabled Safety-Critical Scenario Generation for Autonomous Vehicles (CVPR 2024 · similarity 0.84)
- Flow Matching-Based Autonomous Driving Planning with Advanced Interactive Behavior Modeling (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
