mlsanityData to 14 Oct 2026

NavSafe-∞: Benchmarking Closed-Loop Driving Safety in Photorealistic Environments

arXiv 2609.26618 ↗First posted 22 Sep 2026cs.ROcited by 0

On the map: End-to-end autonomous driving → Adversarial scenarios and safety testing

Abstract (arXiv)

End-to-end (E2E) driving policies have advanced rapidly on open-loop (OL) benchmarks, yet OL evaluation cannot reveal whether a policy can withstand compounding errors, recover from failures, or interact safely with surrounding actors. We introduce NavSafe-∞, a photorealistic closed-loop (CL) benchmark comprising 280 scenarios spanning 28 event types, each with success and failure criteria defined within a structured traffic-safety taxonomy, yielding category-level capability scores for Traffic Crashes, Vulnerable Road User Crashes, Traffic Violations, and Traffic Incidents. After evaluating 20 E2E policies, we find that OL gains do not reliably transfer to CL safety. Analysis of two common remedies reveals that (1) passive demonstration perturbation helps mainly when CL rollouts stay near their perturbed training states, and (2) OL reinforcement-learning fine-tuning exhibits reward hacking by trading safety margin for ego progress, which CL feedback amplifies into compounding safety-critical errors. Together, these results demonstrate the blind spot of OL benchmarks indicating CL safety success. The benchmark and an extensible toolbox for customizable event curation and policy diagnosis will be open-sourced to facilitate future research.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Adversarial scenarios and safety testing, one of 9 directions in End-to-end autonomous driving. Builds on 50 of its 57 references that are in the map. First posted on arXiv 22 Sep 2026.

Builds on

50 of 57 references.

Closest ancestors · 2

Same direction, newest first

  1. 2026HiDrive: A Closed-Loop Benchmark for High-Level Autonomous Driving11 May
  2. 2025Pseudo-Simulation for Autonomous Driving4 Jun

Other references · 48

  1. 20233D Gaussian Splatting for Real-Time Radiance Field Renderingcited by 6,226
  2. 2015Rethinking the Inception Architecture for Computer Visioncited by 5,475
  3. 2019nuScenes: A multimodal dataset for autonomous drivingcited by 4,523
  4. 2017Quo Vadis, Action Recognition? A New Model and the Kinetics Datasetcited by 3,270
  5. 2025Wan: Open and Advanced Large-Scale Video Generative Modelscited by 2,708
  6. 2019Scalability in Perception for Autonomous Driving: Waymo Open Datasetcited by 2,451
  7. 2017CARLA: An Open Urban Driving Simulatorcited by 2,449
  8. 2018Towards Accurate Generative Models of Video: A New Metric & Challengescited by 1,234
  9. 2024Emu3: Next-Token Prediction is All You Needcited by 781
  10. 2021NuPlan: A closed-loop ML-based planning benchmark for autonomous vehiclescited by 493
  11. 2024NAVSIM: Data-Driven Non-Reactive Autonomous Vehicle Simulation and Benchmarkingcited by 354
  12. 2024DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Drivingcited by 334
36 more
  1. 20163D Menagerie: Modeling the 3D shape and pose of animalscited by 312
  2. 2024tinyBenchmarks: evaluating LLMs with fewer examplescited by 310
  3. 2021MetaDrive: Composing Diverse Driving Scenarios for Generalizable Reinforcement Learningcited by 290
  4. 2025AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuningcited by 260
  5. 2024Bench2Drive: Towards Multi-Ability Benchmarking of Closed-Loop End-To-End Autonomous Drivingcited by 252
  6. 2023Parting with Misconceptions about Learning-based Vehicle Motion Planningcited by 242
  7. 2019CenterMask : Real-Time Anchor-Free Instance Segmentationcited by 201
  8. 2025ReCogDrive: A Reinforced Cognitive Framework for End-to-End Autonomous Drivingcited by 198
  9. 2025DriveVLA-W0: World Models Amplify Data Scaling Law in Autonomous Drivingcited by 134
  10. 2024DriveArena: A Closed-loop Generative Simulation Platform for Autonomous Drivingcited by 98
  11. 2024HUGSIM: A Real-Time, Photo-Realistic and Closed-Loop Simulator for Autonomous Drivingcited by 90
  12. 2025Generalized Trajectory Scoring for End-to-end Multimodal Planningcited by 67
  13. 2025DriveLaW:Unifying Planning and Video Generation in a Latent Driving Worldcited by 47
  14. 2026Driving on Registerscited by 44
  15. 2025RAP: 3D Rasterization Augmented End-to-End Planningcited by 41
  16. 2026Kimodo: Scaling Controllable Human Motion Generationcited by 37
  17. 2025SimScale: Learning to Drive via Real-World Simulation at Scalecited by 36
  18. 2026SparseDriveV2: Scoring is All You Need for End-to-End Autonomous Drivingcited by 26
  19. 2024WOMD-Reasoning: A Large-Scale Dataset for Interaction Reasoning in Drivingcited by 19
  20. 2026Fail2Drive: Benchmarking Closed-Loop Driving Generalizationcited by 17
  21. 2026SpanVLA: Efficient Action Bridging and Learning from Negative-Recovery Samples for Vision-Language-Action Modelcited by 17
  22. 2026NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle Simulationcited by 14
  23. 2026DiffusionHarmonizer: Bridging Neural Reconstruction and Photorealistic Simulation with Online Diffusion Enhancercited by 10
  24. 2026BridgeSim: Unveiling the OL-CL Gap in End-to-End Autonomous Drivingcited by 9
  25. 2026nuReasoning: A Reasoning-Centric Dataset and Benchmark for Long-Tail Autonomous Drivingcited by 9
  26. 2026SimWAM: A Simple World Action Model for End-to-End Autonomous Drivingcited by 9
  27. 2026Beyond Imitation: Learning Safe End-to-End Autonomous Driving from Hard Negativescited by 8
  28. 2026World Engine: Towards the Era of Post-Training for Autonomous Drivingcited by 8
  29. 2026SOMA: Unifying Parametric Human Body Modelscited by 6
  30. 2026Do Open-Loop Metrics Predict Closed-Loop Driving? A Cross-Benchmark Correlation Study of NAVSIM and Bench2Drivecited by 6
  31. 2026Instant NuRec: Feed-Forward 3D Gaussian Reconstruction for Driving Scene Simulationcited by 3
  32. 2025Controllable Motion Generation via Diffusion Modal Couplingcited by 2
  33. 2026MTDrive: Multi-turn Interactive Reinforcement Learning for Autonomous Drivingcited by 2
  34. 2026Understanding Real-World Traffic Safety through RoadSafe365 Benchmarkcited by 2
  35. 2026Asset Harvester: Extracting 3D Assets from Autonomous Driving Logs for Simulationcited by 2
  36. 2026Post-Training in End-to-End Autonomous Drivingcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026AutoAgent0: An Agentic Runtime for Safe Closed-Loop Driving
  2. 2026DriveSafer: End-to-End Autonomous Driving with Safety Guidance
  3. 2026RefineDrive: Reliable Failure-Guided Learning for Vision-Language-Action Driving
  4. 2025RoaD: Rollouts as Demonstrations for Closed-Loop Supervised Fine-Tuning of Autonomous Driving Policies
  5. 2026CounterPlay: Counterfactual Post-Training for Self-Play Driving Policies
  6. 2025AD-R1: Closed-Loop Reinforcement Learning for End-to-End Autonomous Driving with Impartial World Models
  7. 2026Closed-Loop Refinement and Execution for Learned Driving Planners
  8. 2024NeuroNCAP: Photorealistic Closed-loop Safety Testing for Autonomous Driving