[Feature]: robot.set_qpos and robot.control_dofs_position have different indexing

[mhwang003]

What feature or enhancement are you proposing?

HI, Thanks for the incredible work!

Using Genesis has been exciting and quite intuitive.

I noticed when using robot.set_qpos and robot.control_dofs_position for the go2 robot, the indexing for the joints was different.
for robot.set_qpos the indexing was

FL_indices = torch.tensor([0, 4, 8]) FR_indices = torch.tensor([1, 5, 9]) RL_indices = torch.tensor([2, 6, 10]) RR_indices = torch.tensor([3, 7, 11])
and for robot.control_dofs_position it was

FR_indices_target = torch.tensor([0, 1, 2]) FL_indices_target = torch.tensor([3, 4, 5]) RR_indices_target = torch.tensor([6, 7, 8]) RL_indices_target = torch.tensor([9, 10, 11])

Am I doing something wrong? or If possible having the same indexing would be awesome. Thanks

Motivation

easier indexing

Potential Benefit

less bugs

What is the expected outcome of the implementation work?

same indexing for the 2 methods

Additional information

No response

[zswang666]

qpos and dofs are different concepts and thus they won't necessarily share the same indexing. Specifically, for free joints, qpos has 7 dimensions and dofs has 6. But the actual indexing you provided looks quite weird indeed. Could you provide a simple script that show how you arrived at your observation?