[Question] Not converge when action is zero

[jaykorea]

Question

Thanks for your well made project!
Well, I've tested wheeled env and transfered it to mujoco to validate the policy. However I have no idea why does the robot keep slightly move foward evne if there is no cmd velocity. Is this because of isaacsim precision issue?

Thanks

[fan-ziqi]

Adjusting the reward function can solve this problem, such as penalizing joint movements at 0cmd