Show HN: 178K Parameter Neural Net That Wins Poke(rogue)like

Category: ai-ml

Tags: reinforcement-learning, pokemon, game-ai

Score: 6.8/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 4)

This project creates a reinforcement learning agent (PPO) with only 178K parameters that learns to beat a Pokémon roguelike game by hooking into its real client-side JavaScript logic. It stands out for its headless simulator built directly from the game's code, enabling fast training and direct transfer to the live browser environment, though its utility is limited to this specific game.

Target audience: AI researchers, game developers

Repository: https://blog.thiagolira.com.br/i-got-so-mad-at-poke-rogue-like-that-i-trained-a-rl-agent-to-beat-it-for-me · JavaScript · NOASSERTION

View on Hacker News