Kaiyang Chen

“This paper solves a specific, painful failure mode in deployed humanoid systems — robots that confidently declare task completion when they haven't actually succeeded — by building a geometry-grounded verification layer that sits between the VLM planner and the physical controller.”