Demo-trained policies failed at deployment. So they let an operator grab the arms mid-run and retrained on the corrections.
I3L, from @wensi_ai at Stanford. At least 90% on all 8 tasks.
I3L, from @wensi_ai at Stanford. At least 90% on all 8 tasks.
0:58