Robot-Use LLM Agents
Robot-Use LLM Agent and VLA/VLM Safety
Safety research for robot-use LLM agents and VLA/VLM systems.
A robot driven by an LLM, a VLA, or a VLM acts in the physical world, next to people. We research how to make that safe.
To understand the risk, we study two questions. First, what kind of AI makes people lower their guard and feel closer to it, even without noticing. Second, whether such an AI will do something that harms a person, especially when it can operate a machine, and whether it notices the harm and stops once it has begun.
Branches
Harm
Whether a model that can operate a machine will harm a person, and whether it notices and stops.
Trust
What kind of AI makes people lower their guard.