Show HN: Play an RL self-play policy in Pokemon Gen 3 Randbats in the browser

Zachary Kingston, a Purdue University researcher, released a browser-based demo that lets players battle Pokémon Showdown's Gen 3 Random Battle format against an AI policy trained via reinforcement learning self-play on a massively parallel GPU battle engine built to exactly match Showdown's simulator. The demo runs entirely on the user's device, and the AI does not cheat, guessing only from what it has observed of the player's team. The project is built on Pokémon Showdown 0.11.11 by Guangcong Luo and contributors under the MIT License and is a non-commercial fan project unaffiliated with Nintendo, Game Freak, or Creatures Inc.

Pokémon Showdown's Gen 3 Random Battle, played in your browser against AI trained with a massively parallel GPU battle engine built to exactly match Showdown's simulator. Everything runs on your device The AI is also not cheating, it only guesses based upon what it has seen of your team. Battle simulator, game data, and move, ability and item text: Pokémon Showdown 0.11.11 by Guangcong Luo and contributors MIT License . Sprites: the games', from Showdown's sprite server. Pokémon and all related names and images are © Nintendo, Game Freak and Creatures Inc. This is a non-commercial fan project, not affiliated with or endorsed by them. Made by Zachary Kingston zkingston@purdue.edu with Claude. Code on Github. Loading the simulator and network…