stable-retro

Reinforcement learning library with support for pcsx2 and opengl, among other cores.

Comparing Behaviour Trees and Deep Reinforcement Learning for Fighting AI

Recurrent PPO + RND reinforcement-learning engine with vectorized environments, resumable checkpoints, and telemetry.

PPO reinforcement-learning agents that learn to play classic games from scratch - DOOM, Pokemon Red, and Street Fighter II - one standalone trainer per branch. Experimental R&D; bring your own ROMs.

In-house Mortal Kombat tournament environment based on stable-retro

Cookiecutter scaffold for reproducible Stable Retro reinforcement-learning projects with tested project generation and CI.

RL environment for the 2005 GBA game 'Hot Wheels Stunt Track Challenge'

An agent that learns to play NES games from pixels and sound alone — the emulator's memory is off limits to the policy. Motion-based object tracking, on-screen counter reading, world model, planner, cross-game transfer.

Reinforcement learning agent for Street Fighter II using PPO, trained with stable-retro and Stable-Baselines3

A reinforcement learning A3C implementation trained to play Super Mario Bros

Deep Q-Networks for Super Mario Bros 3, from a beginner-friendly basic implementation to advanced Rainbow-style improvements.

Train a PPO agent to play Super Mario Bros using Stable-Baselines3 on stable-retro.

PPO agents for games