IVRA: Improving Visual-Token Relations for Robot Action Policy with Training-Free Hint-Based Guidance
BeginnerJongwoo Park, Kanchana Ranasinghe et al.Jan 22arXiv
IVRA is a simple, training-free add-on that helps robot brains keep the 2D shape of pictures while following language instructions.
#Vision-Language-Action#affinity map#training-free guidance