Thanks for the work. Currently the obs has attribute gripper_pose in the world frame. I would like to do imitation learning with a model's action space defined as the delta gripper pose. I wrote a transformation function by myself, which use the next obs.gripper_pose and the current one. Transfer them from world frame to the end effector frame, then do reduction. But I found it is not perfect and I do not know reason. Could you please give me some insight?
Hi,
Thanks for the work. Currently the obs has attribute gripper_pose in the world frame. I would like to do imitation learning with a model's action space defined as the delta gripper pose. I wrote a transformation function by myself, which use the next obs.gripper_pose and the current one. Transfer them from world frame to the end effector frame, then do reduction. But I found it is not perfect and I do not know reason. Could you please give me some insight?