Closed stuartcrobinson closed 2 years ago
One possible option that you never send done=True. So when rlg tries to save model there is no final rewards. Ill fix it on my side by checking that there is no mean_rewards. But there is possible error in env.
Hello. I've run into the same issue, also working with Isaac Gym (IsaacGymEvs specifically). @Denys88 it is not clear to me what you mean with:
you never send done=True.
Can you please advice further on how this issue may be resolved?
Hi @yorgosk here are two options: 1) I have 'save_best_after' parameter which is defaulted to 100. If you never send done = True in any of your envs for the first 100 epochs you will get exact this error. if you set it to the very high value issue should disappear if everything is fine with your env. 2) you never send done == True because of the errors in your code. If episode never ends I cannot return statistics.
(this is on line 1214 in the version that isaac gym is using):
https://github.com/Denys88/rl_games/blob/a33b6c4d4012935e6f494508e1992f4eda4f8104/rl_games/common/a2c_common.py#L1207
the only changes i made in Isaac Gym is this function in
isaacgymenvs/tasks/cartpole.py
:error:
hope this is helpful. i'm new to this stuff