VLS: Steering Pretrained Robot Policies via Vision-Language Models
IntermediateShuo Liu, Ishneet Sukhvinder Singh et al.Feb 3arXiv
Robots often learn good hand motions during training but get confused when the scene or the instructions change at test time, even a little bit.
#Vision–Language Steering#Inference-time control#Diffusion policy