Reproduce “DanZero: Mastering GuanDan Game with Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yudong Lu, Jian Zhao, Youpeng Zhao, Wengang Zhou, Houqiang Li
DanZero applies self-play reinforcement learning to GuanDan, scoring legal card plays with a neural network trained from completed-game outcomes. It represents cards played, teammates, and wild cards, while using heuristics for tribute decisions. After 30 days, it won 87.45% of 1,000-game comparisons against baseline 4, but this result does not establish superiority over every opponent.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.