Cooperative Open-ended Learning Framework for Zero-shot Coordination

• DOI: 10.48550/arxiv.2302.04831
• Publication Date: 2023-01-01
• Authors: Yang Li; Zhang Shao; Jichen Sun; Yali Du; Ying Wen; Xinbing Wang; Wei Pan
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Machine Learning (cs.LG)

Zero-shot coordination in cooperative artificial intelligence (AI) remains a significant challenge, which means effectively coordinating with wide range of unseen partners. Previous algorithms have attempted to address this challenge by optimizing fixed objectives within population improve strategy or behaviour diversity. However, these approaches can result loss learning and an inability cooperate certain strategies the population, known as incompatibility. To issue, we propose Cooperative Open-ended LEarning (COLE) framework, constructs open-ended games two players from perspective graph theory assess identify ability each strategy. We further specify framework practical algorithm that leverages knowledge game theory. Furthermore, analysis process shows it efficiently overcome The experimental results Overcooked environment demonstrate our method outperforms current state-of-the-art methods when different-level Our demo is available at https://sites.google.com/view/cole-2023.