Make arm MIT control fully accessible#8
Conversation
kou
changed the title
| self.last_command = target_pos | ||
| Args: | ||
| position: Desired joint positions for the mit motors (first n). | ||
| velocity: Desired joint velocity feedforward. Defaults to zeros if None. |
There was a problem hiding this comment.
Should we add the _ff suffix to velocity like we did for torque_ff?
There was a problem hiding this comment.
Usually it is only added to feed forward torque term. The velocity comment is inaccurate and I will fix that.
| velocity: Desired joint velocity feedforward. Defaults to zeros if None. | ||
| torque_ff: Desired joint torque feedforward. Defaults to zeros if None. |
There was a problem hiding this comment.
I'm not familiar with MIT control but is it natural that we use zeros as the default value? If zeros isn't so natural, it may be better that we require velocity and torque instead of making them optional.
There was a problem hiding this comment.
Having zeros at these places approximate position control (which is commonly used in current policy deployments). I agree that we can require them for now. However, I am wondering if we should make the gains (kp/kd) also accessible @tokirobot any inputs for this?
Co-authored-by: Sutou Kouhei <kou@cozmixng.org>
2 participants
The goal is to make the arm motors’ MIT control fully accessible. Right now, the
send_positionmethod sets velocity and torque to zero and doesn’t expose those parameters, but many use cases require them. At the same time, the functionality withsend_positionhas been preserved.However, the current index management approach with mixed control mode (using self.num_mit_motors) feels somewhat fragile and error-prone. Appreciate more input thoughts on this.