Speed deviation

[nepenthes123]

Hello, have you also encountered the issue where the strategy trained based on the code provided by the project moves the robot even when the initial angular velocity and linear velocity are set to 0 when simulating in Mujoco?

[kf953741]

Me too, I have the same question

[hang0610]

Hi, I suggest you to add symmetric data augmentation to make the policy output symmetric action, which can solve this issue. You can check how holosoma does for this, which I contributed.