HazardArena shows VLA models trained on safe data frequently produce unsafe actions in semantically risky but visually similar settings, and a training-free Safety Option Layer reduces those failures with little performance cost.
ISO 13482:2014 robots and robotic devices — safety requirements for personal care robots
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
HazardArena: Evaluating Semantic Safety in Vision-Language-Action Models
HazardArena shows VLA models trained on safe data frequently produce unsafe actions in semantically risky but visually similar settings, and a training-free Safety Option Layer reduces those failures with little performance cost.