提出机器人首次社交动作需获得授权，填补通用机器人安全空白。
Initiation Safety: A Missing Dimension in Generalist-Robot Safety
- 引入‘启动授权’概念，判断机器人是否该发起初次不可逆社交行为。
- 实验证明，先探测再授权比直接行动更安全，降低不当互动风险。
- 适合研究人机交互安全、通用机器人伦理的学者与工程师参考。
通用机器人安全通常聚焦于运动或对话层面。我们指出一个被忽视的维度：机器人是否应发起首个难以撤销的社交行为，如问候、未经邀请的抓握或进入他人空间？我们称此为‘启动授权’。当前框架很少将其视为独立的安全层。现有系统常跳过此步骤，仅凭高参与度评分或自信的视觉语言模型（VLA）输出就允许行动。但看到人并不等于获得对方同意。本文将启动授权纳入通用机器人安全框架，对比其与事后计划的VLA防护机制，在门廊人形机器人上实现‘探测-授权-发声’（PAS）流程，并在记录数据上与直接发起（direct-init）对比。此外，设计了三条件用户研究，提出关于评估指标、治理机制以及启动授权与基础模型生成边界等开放问题。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。