analiza aiqbench
EgoSafetyBench introduces a rigorous diagnostic tool for evaluating vision-language models acting as runtime safety guards for embodied agents in real-world environments like homes and factories. The benchmark reveals critical vulnerabilities where models fail to pinpoint specific hazard moments or are misled by superficial cues, highlighting that apparent robustness often stems from indiscriminate intervention rather than true physical reasoning. This work is essential for advancing the reliability of AI-driven safety systems before their deployment in sensitive operational settings.
EgoSafetyBench: A Diagnostic Egocentric Video Benchmark for Evaluating Embodied VLMs as Runtime Safety Guards
arXiv:2607.00218v2 Announce Type: replace-cross Abstract: Vision-language models (VLMs) are increasingly proposed as runtime safety guards for embodied agents in homes and factories. A deployable guard must catch genuinely unsafe situations while avoiding unnecessary intervention on routine but superficially alarming activity, a distinction obscured by binary safety benchmarks. We introduce EgoSafetyBench, an egocentric video benchmark of 1,200 robot-view scenarios annotated at half-second granularity, with two evaluation tracks. The situational track (800 scenarios) spans routine, safe-but-suspicious, obvious-hazard, and contextual-hazard scenes. The visual-channel track (400 scenarios) tests whether misleading in-scene text corrupts physical-safety judgments, using matched truthful controls. Both tracks use contrastive ladders: near-identical scenarios differing in a single visible deciding cue, forcing predictions to hinge on that cue. Across ten open- and closed-source VLMs, we find that guards often recognize videos containing hazards yet miss the specific hazardous moments, especially for contextual hazards. Misleading in-scene signs further degrade all tested guards: vulnerable models miss up to a third of hazards, while seemingly robust models often over-intervene on safe content. Matched controls show that apparent robustness can reflect indiscriminate alarming rather than true physical reasoning. A 20-clip real-video sanity check further shows that model rankings transfer beyond synthetic rendering, with Spearman \r{ho} = 0.87 for hazard miss rate and 0.94 for visual-channel mismatch recall.
Abstract:Vision-language models (VLMs) are increasingly proposed as runtime safety guards for embodied agents in homes and factories. A deployable guard must catch genuinely unsafe situations while avoiding unnecessary intervention on routine but superficially alarming activity, a distinction obscured by binary safety benchmarks. We introduce EgoSafetyBench, an egocentric video benchmark of 1,200 robot-view scenarios annotated at half-second granularity, with two evaluation tracks. The situational track (800 scenarios) spans routine, safe-but-suspicious, obvious-hazard, and contextual-hazard scenes. The visual-channel track (400 scenarios) tests whether misleading in-scene text corrupts physical-safety judgments, using matched truthful controls. Both tracks use contrastive ladders: near-identical scenarios differing in a single visible deciding cue, forcing predictions to hinge on that cue. Across ten open- and closed-source VLMs, we find that guards often recognize videos containing hazards yet miss the specific hazardous moments, especially for contextual hazards. Misleading in-scene signs further degrade all tested guards: vulnerable models miss up to a third of hazards, while seemingly robust models often over-intervene on safe content. Matched controls show that apparent robustness can reflect indiscriminate alarming rather than true physical reasoning. A 20-clip real-video sanity check further shows that model rankings transfer beyond synthetic rendering, with Spearman \r{ho} = 0.87 for hazard miss rate and 0.94 for visual-channel mismatch recall.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2607.00218 [cs.CV]
(or arXiv:2607.00218v2 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2607.00218 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 304 cuvinte
Subiecte conexe
NVIDIA Releases Cosmos 3: A Two-Tower Mixture-of-Transformers Foundation Model Unifying Physical Reasoning, World Generation, and Action Generation
MarkTechPost · 3 iun. 2026
Planning Takes More Than Token Prediction: Causal Plan for Benchmarking and Building Physically Grounded Embodied Reasoners
arXiv · 25 sept. 2026
Safe Task Planning with Long-Term Graph Memory for Embodied Agents
arXiv · 9 sept. 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · 25 sept. 2026