Pinned
What if robot hands could acquire dexterity once, then reuse it to learn downstream tasks much faster?
Introducing ADEPT: a pre-training & post-training paradigm using RL entirely in sim, then zero-shot deploying visuo-tactile policies in the real world.
