huggingface
Question/Potential Bug about SO100 use_degrees mode for sim2real/real2sim alignment
Hi all, I just saw that a new hardware API / setup has been merged. It is definitely cleaner than before and appreciate the efforts!
However there are some changes that make things possibly breaking for any use case with sim2real or real2sim research/work.
One of them is that the SO100 follower robot defaults to -100 to 100 or 0 to 100 normalization modes for the joints. This means that simulation related code must convert the normalized readings to degrees by reading the calibration file, which gets quite messy as the calibration file really should only be handled by lerobot.
I notice that the SO100 and other robots support an use_degrees configuration option which alleivates this issue to an extent. This makes the hardware read joint angles now in a range/scale that can be aligned with simulation, however this option does not affect the gripper readings, which are still 0 to 100 and misaligned.
Would it be possible to make it such that in the use_degrees mode the gripper is also changed to use that mode as well?
Hey @StoneT2000
What would be the best practice in your opinion in this case?
The current MotorNormMode.DEGREES would not make sense for the gripper as the zero position would be midway between opening and closing the gripper. We have generally used the gripper as a linear joint.
We could add the possibility to use degrees for gripper and shift the zero position to be in close state.
Or should we add a linear joint in the urdf?
Fixed now with some of the new URDFs. Issue here can at least help others be aware if they try sim2real as well!