Initiation Safety: A Missing Dimension in Generalist-Robot Safety
arXiv:2607.07420
The paper introduces "initiation authorization" as a missing safety layer for generalist robots, focusing on whether a robot should take its first irreversible social action (e.g., greeting or reaching) without explicit consent, and proposes a probe‑authorize‑speak framework to evaluate this concept.
Abstract
Safety for generalist robots is usually discussed in terms of motion or dialogue. We argue a third question is missing: should the robot take its first hard-to-undo social action at all, such as a greeting, an uninvited grasp, or stepping into someone's space? We call this initiation authorization. Current frameworks rarely treat it as a separate safety layer. Today's stacks often skip this step: a high engagement score or a confident VLA rollout is treated as permission to act. But seeing a person is not the same as having their consent to be addressed. We frame initiation authorization within generalist-robot safety and contrast it with post-plan VLA guardrails, implementing PAS (probe-authorize-speak) on a doorway humanoid, comparing it with direct-init on logged traces, and proposing a three-condition user study, with open questions on metrics, governance, and where initiation ends and foundation-model generation begins.
4 pages, 2 figures. Accepted to RSS 2026 Workshop on Rethinking Safety for Generalist Robots