How to use different robotic arms and grippers such as the UR robotic arm to make your own datasets

[haicaoha]

i do not have panda robotic arm?Can i realize the mimic play in UR robotic arm?If yes, how to make my own datasets for the specific task?

[j96w]

Hi @haicaoha , this repo is not specific to the robot hardware choices. As long as you have an end-effector_position_controller of the UR, it works in the same way as the Franka and the dataset format will also be the same. For your own datasets, you can follow this script to use a spacemouse to control the UR (with end-effector_position_control) for data collection.

[haicaoha]

I mean making a dataset for other scenarios and not kitchen scenes， I found out that bddl file exists in LEBERO library, how do I make my own simulation scene and BDDL file?

[j96w]

I see. It might be better to directly ask the author of LIBERO in their repository to build or modify the LIBERO scene. We didn't participate in the implementation of LIBERO. To modify a BDDL file, you can edit the symbolic predicates to achieve the goal you want. The environment will then directly convert these predicates into a reward function to check task success.

[haicaoha]

Thank you for your answer, I've completed the training in the simulation environment. But if I want to use human data to train in a real environment, how can I achieve inference after obtaining the training weight file【pth】, that is, really control the robotic arm to complete a specific task?

[haicaoha]

and could you share the code about the training process in the real environment