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

arXiv: https://arxiv.org/abs/2606.14238 · HTML: https://mlsanity.com/paper/2606.14238/

## 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.

## Placed in

- End-to-end autonomous driving / [Reasoning VLA and VLM driving policies](https://mlsanity.com/driving/reasoning-vla-and-vlm-driving-policies/)

## Builds on (most cited first, 10 of 10)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control](https://mlsanity.com/paper/2307.15818/) (2023)
- [Senna: Bridging Large Vision-Language Models and End-to-End Autonomous Driving](https://mlsanity.com/paper/2410.22313/) (2024)
- [Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail](https://mlsanity.com/paper/2511.00088/) (2025)
- [Are VLMs Ready for Autonomous Driving? An Empirical Study from the Reliability, Data, and Metric Perspectives](https://mlsanity.com/paper/2501.04003/) (2025)
- [Emergence of Separable Manifolds in Deep Language Representations](https://mlsanity.com/paper/2006.01095/) (2020)
- [Visual Adversarial Attack on Vision-Language Models for Autonomous Driving](https://mlsanity.com/paper/2411.18275/) (2024)
- [The RoboDrive Challenge: Drive Anytime Anywhere in Any Condition](https://mlsanity.com/paper/2405.08816/) (2024)
- [RoboDriveVLM: A Novel Benchmark and Baseline towards Robust Vision-Language Models for Autonomous Driving](https://mlsanity.com/paper/2512.01300/) (2025)
- [Lost in Fog: Sensor Perturbations Expose Reasoning Fragility in Driving VLAs](https://mlsanity.com/paper/2605.21446/) (2026)

## 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.