PantheonRL is a package for training and testing multi-agent reinforcement learning environments. PantheonRL supports cross-play, fine-tuning, ad-hoc coordination, and more.
Instead of
abs_pos_p1 = np.array(p0.position)
it should be
abs_pos_p1 = np.array(p1.position)
This needs to be refactored to wrap a more recent version of that overcooked_ai repo where the feature array typo was fixed. Or this file needs to be corrected after cloning the old repo.
The PantheonRL environment wrapper for overcookedgym uses an old version of HumanCompatibleAI's overcooked_ai repo under the hood. Unfortunately, this old version contains a typo in the p1 feature array here: https://github.com/HumanCompatibleAI/overcooked_ai/blob/6eaceb0a9a2501f1b9fccbf4c7016d6662ed1108/overcooked_ai_py/mdp/overcooked_mdp.py#L1157
Instead of
abs_pos_p1 = np.array(p0.position)it should beabs_pos_p1 = np.array(p1.position)This needs to be refactored to wrap a more recent version of that overcooked_ai repo where the feature array typo was fixed. Or this file needs to be corrected after cloning the old repo.