Confusion about the admittance control

[zichunxx]

Hi! Thanks for your great sharing!

I'm curious about the implementation of the admittance control that if it is necessary to compensate for the gravity of the gripper to require precise force sensor data.

Thanks!

[mdrolet01]

Hi @zichunxx, a gravity term is included in our OSC class and we apply the rotation transformation for the gripper's coordinate frame here: https://github.com/ir-lab/bimanual-imitation/blob/main/irl_control/device.py#L202. If you notice the gripper are pulled down when the admittance gain is increased, this is due to the forces of the object. To avoid making the controller too specific, you can decrease/increase the gain depending on the object/task.

[zichunxx]

Thanks for your kind reply!

If you notice the gripper are pulled down when the admittance gain is increased, this is due to the forces of the object. To avoid making the controller too specific, you can decrease/increase the gain depending on the object/task.

I will try to tune the admittance gain to neutralize the object force.

I'm unfamiliar with the admittance control and don't know if the admittance control,

i.e.,

https://github.com/ir-lab/bimanual-imitation/blob/2676916f7f0092697e04b67986c5d2e1b2a39f85/irl_control/osc.py#L243-L246

significantly impacts insertion success. Will the admittance control perform better than the pure OSC mode?

Looking forward to getting some of your expertise!

[mdrolet01]

For this task, you may just want to use a small gain (such as the default one). A small admittance term can in theory help cope with the contact forces during the peg insertion. But the extent to which this improves the performance might be marginal.

[zichunxx]

Thanks for your help! @mdrolet01

[mdrolet01]

No problem! :)