Merlin Human-AI Teaming Toolbox

This library implements the Any-Play learning augmentation in Hanabi Learning Environment. Any-Play is an intrinsically motivated, diversity-based augmentation for reinforcement learning algorithms (RL) that enables RL agents to effectively cooperate with novel, never-before-seen teammates on collaborative tasks, referred to as zero-shot coordination. This library is used to train and demonstrate such teaming in the cooperative card game Hanabi; although the Any-Play augmentation is environment agnostic and, therefore, could be applied to many other domains beyone Hanabi.

This library, "hanabi_AnyPlay", is a derivative of the hanabi_SAD library by FACEBOOK, used under the CC BY-NC license license. "hanabi_AnyPlay" is licensed under the same CC BY-NC license by MIT.