Go1 Support

[lucasbrodo]

Hi @silvery107, Thanks for this very rich work.

Could you list some guidelines for deploying your code to Unitree Go1?

Thanks for the feedback.

Lucas

[silvery107]

Since the MPC controller and the RL training environment are decoupled, you can deploy the trained policy network using RobotRunnerPolicy, and you will need to connect its interface to unitree sdk for sending motor commands and reading sensor data.

[lucasbrodo]

Hi @silvery107, I have connected the unitree SDK interface for sending motor torques commands and reading sensor data. I have only modified the RL_MPC_Locomotion.py and I'm just focusing on getting RobotRunnerFSM running for now. I don't know exactly where to modify the script for providing the real robot foot forces/contacts since it's vital for the MPC.

Thank you very much for your help.

[silvery107]

What do you mean by providing the foot forces/contacts?

Please note that this codebase is not designed to run on a real robot (the sim2real part is not included), so run it with extreme caution and at your own risk.

[lucasbrodo]

Sorry for the confusion, I thought that the foot contacts were used for the MPC controller but it appears that they are predicted (if I refer to the original MIT-Cheetah paper).

Could you indicate what is needed specifically for the deployment on the real robot ? Which parameters/values need to be modified ?

In my case, the controller outputs very high torques after providing the sensors' values from the real robot. The robot can't even stand up.

Again, thanks for your help.

[silvery107]

For parameter tuning, try modify the joint Kp, Kd and weights of MPC solver.

Also a filter (e.g. EKF) is needed for sensor data, since the current state estimator is basically a dummy version.

[silvery107]

Closed due to inactivity