Action Space Representation and URDF

[oxFFFF-Q]

Thank you for your excellent work on this project. I believe there are many exciting possibilities for further exploration in this area.

I have a question regarding the choice of action space representation. Could you please explain why you chose joint angles as the action space instead of end-effector poses? I believe the latter offers greater generalization.

Additionally, I am interested in testing the generated joint angles by converting them to end-effector poses for use on other robots. Have you attempted this approach?

Since calculating a robot's forward kinematics requires a URDF file, could you provide the URDF file or a link to it for the same robot model used in your project?

Thank you very much!

[pratikgehlott]

follow

[Kelym]

Hope the answer here helps.

TL;DR: Joint position space was the most convenient choice at the time (since all robots have it by default). But we later (after pi0) tried experiments with end effector poses as well.