Initiation Safety: A Missing Dimension in Generalist-Robot Safety

2026-07-08

AI Summary

This work addresses a critical gap in existing general-purpose robot safety frameworks: the lack of consideration for “initiation authorization”—whether a robot should perform potentially irreversible initial social actions, such as unsolicited greetings, unpermitted grasping, or intrusions into personal space. The paper establishes initiation authorization as a distinct safety dimension separate from motion and dialogue safety and introduces a Probe-Authorize-Act (PAS) mechanism that integrates a vision-language-action model, a behavior detection module, and a user authorization interface to enable context-aware initiation decisions. Real-world evaluation on a hallway-deployed humanoid robot demonstrates that PAS significantly outperforms direct initiation strategies, and preliminary user studies confirm that it better aligns with human expectations regarding social boundaries and consent.

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.

Problem

• initiation safety
• generalist robots
• initiation authorization
• robot safety
• social interaction

Innovation

• initiation authorization
• generalist robot safety
• probe-authorize-speak
• social action
• VLA guardrails