Action Accurancy in OSC_POSE controller

[williampai0704]

Hi,
thanks for the work of this amazing real-time library, I was trying to replay my eef delta command using OSC_POSE controller in franka panda and find out that the robot is not moving or weirdly drifting with the input of originally recorded eef delta command.
When I look into the osc_control.py example, I found out that the action_pos is scaled 10 times and clipped between -1 to 1, and the angle command is clipped between -0.5 to 0.5. Also, the default translation scale is 0.05, rotation scale is 1.0, are there any physical meaning or formula behind these parameters?
I try to run osc_control.py and found out the eef does not move as what target_delta_pose specified (I try with moving 0.1m in x but turn out only move around 0.08), so I am wondering is this expected in osc_pose controller.
Thanks for the help.

[yiyang0207]

Same issue here. Is there any update on this problem?

[kingchou007]

Same issue here.

[kingchou007]

I sent a 0.05 delta along +X, but the end-effector undershot by about 13 mm. I use franka research 3.

[jmcoholich]

Same issue. Whenever I send deltas along a single axis (X, Y, or Z) and zeros for rotation, the end-effector rotates while moving along that axis. The translation isn't straight either. Its strange that the Franka cannot achieve accurate operational space commands without a high-level feedback loop.

[jmcoholich]

Update: Increasing translation and rotation Kp values in the controller config helped significantly.