When it comes to strategy changes, the robot may lose control. #71
Description
Hello, we encountered a problem while using Mujoco for simulation. We started the policy node from a single node, which outputs an action based on current state and other information, and then applied this action to the Unitree_Mujoco robot. We have verified these actions on the actual machine and found that they are all feasible. Now, when verifying in Unitree_Mujoco, for some stationary actions, such as squatting and standing up, they can be completed and are consistent with the actual machine. However, for some actions that require movement, such as standing up and then walking, when switching to walking mode, the robot will experience problems like sliding, kicking randomly, and falling over. Moreover, the movements in Mujoco seem to have a lower friction force, and the robot will drift overall? I would like to ask if there are any possible reasons for this?