FarawaySail
commented
2 months ago https://github.com/marlbenchmark/on-policy/issues/104 This may be helpful for your problem. We mainly refer to the original implementation.
Open rickczhao opened 2 months ago
FarawaySail
commented
2 months ago https://github.com/marlbenchmark/on-policy/issues/104 This may be helpful for your problem. We mainly refer to the original implementation.
hello, what a great work you've done! But i found something wrong in onpolicy/onpolicy/runner/shared/grid_runner.py.
in line 524 and 526, variables "obs" and "shared_obs" are actually the same, which are output by function "self._resize_convert" with the same input parameters.However, they are respectively used as the input of the actor and the critic.(maybe the input of the critic should be the global observation?)
Additionally, I find that "obs" and "shared_obs" are not updated in the function "self._resize_convert", in which only raw_ob is updated.