Metamon
paper Your tags
Your notes
Human-level competitive Pokémon (RLC 2025): offline RL over 5M+ human battles reconstructed from Pokémon Showdown replay logs, treating the partially-observed opponent team as a sequence-modeling problem. The resulting agents reached the top 10% of the ranked human ladder — a rare large-scale offline-RL result against real humans. Grigsby and Yuke Zhu, UT-Austin-RPL.