One observation: base_step_action_withgrasp(self, action, joint_name="moving_side") - the joint_name seems to be hardcoded by now and the joint_name not used.
Could be fixed in the future, I do not loose hope to handle diverse arms maybe :)
Then, I observe reward.py seems to have been removed.
I believe in a long-run I might be a good idea to have a proper reward functions library related to re-arrangement tasks.
We could also add penalization like self-collision, velocity constraints and so on.
Thanks @qgallouedec for all the fixes !
One observation: base_step_action_withgrasp(self, action, joint_name="moving_side") - the joint_name seems to be hardcoded by now and the joint_name not used. Could be fixed in the future, I do not loose hope to handle diverse arms maybe :)
Then, I observe reward.py seems to have been removed. I believe in a long-run I might be a good idea to have a proper reward functions library related to re-arrangement tasks. We could also add penalization like self-collision, velocity constraints and so on.