1 paper · 1 filter
Yihao Wu, Jinming Ma, Junbo Tan +5
Diffusion-based policies generalize poorly in semantic manipulation, a key obstacle to real-world deployment, because text-only instructions cannot reliably steer the policy toward…