Mayankm96
commented
1 year ago Hi,
That's an interesting question. Unfortunately, we don't have the resources to check sim-to-real for Unitree robots. However, we encourage contributions from the community on verifying the tuned gains for deployment of policies trained in simulation.
Recently, there has been a work that touches more in detail about tuning controllers for legged locomotion, particularly for the different behaviors you mentioned. I would advise maybe taking a look there. From a quick glance, I saw that they also use an actuator net for the robot, however, the authors of that work would be more resourceful in how useful it is for that particular robot.
Website: https://gmargo11.github.io/walk-these-ways/ Code: https://github.com/Improbable-AI/walk-these-ways
Question
In omni/isaac/orbit/robots/config/unitree.py, the Kp,Kd values of Unitree A1 robot are 25 and 0.5. So what are these values in the reality. Specifically, for various robot behavior (walking,turning, gallop,...), are these parameters uniform (in simulation or reality)? If not, what are the approximate values of each parameter.
Thank you!