Open lightbrighten opened 1 week ago
Hi, I meet same problem. During sim2sim period, the policy trained and performed well can't perform as in play period. I trained a policy to control a robot with more joint, I think maybe I miss somethings to change in mjcf file. Have you solved the problem?
still confused yet, not solved
Keep trying. Once I solved it, I will describe the method here. :)
I haven't solved the problem yet. I would like to share my situation here and see if we have encountered the same problem. After adding other joints and completing training in humanoid Gym, the robot joints in the stage of sim2sim will swing and the robot will slowly fly. I directly added joint and other information to the. ml file in MJCF without modifying the original content, including gravity. I suspect that inertia caused the robot to fly. Do you have any suggestions? (This model is converted by mujoco2.3.6, not which I mentioned above, but the problem is same)
Dear Author Appreciating for your good job!Does this can be used to trainning a method to control a small 10-dof robot,which is 0.45m tall and 10 kgs weight and with "rotate-ankle" joint missing? I have tried this, and the "train" and "play" is stable, but "sim2sim" is unstable, which params I need to adjust in "humanoid_cfg.py"?