mlsanityData to 14 Oct 2026

When and How Severely: Scenario-Specific Safety Envelopes for Driving VLAs

arXiv 2606.14238 ↗First posted 12 Jun 2026cs.RO cs.AIcited by 0

On the map: End-to-end autonomous driving → Reasoning VLA and VLM driving policies

Abstract (arXiv)

Safety certification of Vision-Language-Action (VLA) driving planners under ISO 21448 (SOTIF) rests on an Operational Design Domain (ODD) specification that answers two complementary questions: when does the planner start to fail, and how severely does it fail once it does? We evaluate Alpamayo R1, a 10B-parameter open-weight driving VLA, on 15,968 (clip, attack) pairs. We find a conservative-aggregate gap: an aggregate safe threshold of σ ≤ 50 under a 15% average displacement error (ADE) budget masks well-sampled scenarios that tolerate the top of the tested grid (σ = 70). A Gaussian Mixture Model (GMM) on the changed-explanation subset identifies six discrete severity bands (BIC-optimal k=6), so two perturbation conditions with the same mean error can differ materially in their share of high-severity (C4/C5) failures. Joining the two analyses on the same corpus surfaces a finding neither yields in isolation: the scenarios with the loosest noise thresholds are not those with the lowest high-severity rate: STOPSIGNAL concentrates roughly 4× the C4/C5 share of LANEKEEPING despite tolerating a larger σ. A deployable SOTIF ODD specification for driving VLAs therefore requires a two-dimensional safety envelope, not a single aggregate value per hazard.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reasoning VLA and VLM driving policies, one of 9 directions in End-to-end autonomous driving. Builds on 10 of its 16 references that are in the map. First posted on arXiv 12 Jun 2026.

Builds on

10 of 16 references.

Closest ancestors · 3

Same direction, newest first

  1. 2026Lost in Fog: Sensor Perturbations Expose Reasoning Fragility in Driving VLAs20 May
  2. 2025Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail30 Oct
  3. 2024Senna: Bridging Large Vision-Language Models and End-to-End Autonomous Driving29 Oct

Other references · 7

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  2. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 3,080
  3. 2025Are VLMs Ready for Autonomous Driving? An Empirical Study from the Reliability, Data, and Metric Perspectivescited by 122
  4. 2020Emergence of Separable Manifolds in Deep Language Representationscited by 40
  5. 2024Visual Adversarial Attack on Vision-Language Models for Autonomous Drivingcited by 36
  6. 2024The RoboDrive Challenge: Drive Anytime Anywhere in Any Conditioncited by 23
  7. 2025RoboDriveVLM: A Novel Benchmark and Baseline towards Robust Vision-Language Models for Autonomous Drivingcited by 8

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Localized Conformal Safety Monitoring with Vision-Language Models for Autonomous Driving
  2. 2026DriveSafe: A Hierarchical Risk Taxonomy for Safety-Critical LLM-Based Driving Assistants
  3. 2026Is VLA Reasoning Faithful? Probing Safety of Chain-of-Causation in Autonomous Driving Models
  4. 2026ForesightSafety-VLA: A Unified Diagnostic Safety Benchmark for Vision-Language-Action Models
  5. 2026Revealing Interpretable Failure Modes of VLMs
  6. 2025Evaluation of Safety Cognition Capability in Vision-Language Models for Autonomous Driving
  7. 2026ReasonBreak: Probing Vulnerabilities in Reasoning-Enabled Vision-Language-Action Models for Autonomous Driving
  8. 2026Language2Drive: Do Vision-Language-Action Models Really Follow the Instructions in Autonomous Driving?