About cross-embodiment pretraining #82
Description
Hi, thanks for your impressive work. I learned that you are using EEF pose action mode to mitigate heterogeneous robots difference. While this aligns them into similar semantic action space. But converting Joints States to EEF Pose loses the proprioceptive information. Have you conducted experiments like adding joints into proprioceptive inputs? (Keeps using EEF pose as action label)