Precision regarding the diffIK method

[edwardjjj]

Hi, Thank you for the implementation tutorial.

I'm trying to use the diffIK method to control the eef pose of my robot, but I'm getting a error_norm of 1e-3 (cartesian xyz and quat). This is much higher than what I expect for my use case. Can you please give me some suggestion on improving the precision? Thank you very much!

[fanshJ]

Could it be due to gravity? I just noticed that the gravity compensation in diffik.py and diffik_nullspace.py files is not taking effect. Before the gravity compensation is applied, the value of model.ngravcomp is 0, and after the gravity compensation code like the following, the value of model.ngravcomp remains 0:

if gravity_compensation:
model.body_gravcomp[body_ids] = 1.0

When model.ngravcomp is 0, it seems that MuJoCo does not perform gravity compensation, which could cause the end effector of the robot to be lower than it should be.