Paperopenalex
Beyond Behaviour: A Stimulus-Response Safety Evaluation of Vision-Language-Action Models
Vy Vo, Dinh Phung · arXiv (Cornell University) 2026 · 2026
DOI: 10.48550/arxiv.2610.34837
0citations
Multimodal Machine Learning ApplicationsAdversarial Robustness in Machine LearningRobot Manipulation and LearningAction (physics)
Citations by source
Synced 2026-10-31T02:31:57.837495Z0OpenAlex · primary0Semantic Scholar0Crossref
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Abstract
Humans exhibit biological sensitivity to safety-critical stimuli that can trigger defensive action. Across three models and two video benchmarks, we investigate whether similar responses emerge in vision-language-action models by examining their internal representations and action tendencies in response to visible hazards and harmful consequences. Our findings suggest that task execution may remain the default response even when danger is internally registered.
Related papers
- Are Vision-Language-Action Models Robust to One-Step Observation Perturbations?
arXiv (Cornell University) 2026 · 0 citations
- Do Vision-Language-Action Models Understand Instructions? A Mechanistic Interpretability Study on Language Grounding
arXiv (Cornell University) 2026 · 0 citations
- From Local Evidence to Safety Verdicts: Causal Tracing in Vision-Language Models
arXiv (Cornell University) 2026 · 0 citations
Discussion (0)
No comments yet — start the discussion below.