Question about Pretrain

Thanks for your great work!
During the pretraining stage, is human-centric action data used for training? If so, how is it ensured that the human-centric actions and the robot gripper's actions are in the same space?