Step 1: Teacher in sim (RL/PPO)
A privileged policy sees ground-truth object pose (+ friction, mass, scale) and learns in-hand rotation & SO(3) reorientation. Because it has clean privileged state, it's a much stronger teacher than a proprioception-only one.
A privileged policy sees ground-truth object pose (+ friction, mass, scale) and learns in-hand rotation & SO(3) reorientation. Because it has clean privileged state, it's a much stronger teacher than a proprioception-only one.
1:08 1 2 5
