whereismyai
Pub Street Fighter AI
Pub Street Fighter AI: This is an artificial intelligence reinforcement learning agent that plays Street Fighter II.
This project trains a deep reinforcement learning agent to beat all twelve arcade levels of Street Fighter II Special Champion Edition. It uses a convolutional neural network called NatureCNN and the PPO algorithm to learn movement, attacks, and special moves like the Hadoken and Shoryuken. The model takes processed RGB game screen pixels as input and reads internal memory variables for health to calculate rewards. It supports parallel training across independent emulators and provides final model weights and training logs. The software is built with Python using Gymnasium, Stable-Retro, and Stable-Baselines3 libraries.
Pub Street Fighter AI at a glance
Key features
- Trains on all twelve arcade levels
- Uses PPO reinforcement learning algorithm
- Processes RGB game screen pixels as input
- Supports parallel emulator training
- Provides pre-trained model weights
- Includes TensorBoard logging for training curves
- Features a discrete action space of twenty-seven actions
Use it for
- Train reinforcement learning agents on retro video games
- Test AI model combat performance across multiple levels
- Reproduce training runs with provided logs and weights
Alternatives to Pub Street Fighter AI
- Hangzhou Mahjong — Provides a reproducible tournament protocol, game simulator, and the released r2600 checkpoint.
- AdvSim — Runs AI agents locally in the browser against Pokémon Showdown battle logic.
- AgentGarten — Runs executable code worlds with real-time neural rendering for agents.
- Kungfu — Maintains work state and context when switching between different AI agents.
- AutoTavern — Simulate complete eight-player games and evaluate model judgment in a reproducible environment.
- OmniAgent — It evolves skills in real-time and dynamically hardens security during task execution.
- Rakazo — Complete ownership of data, models, and execution environment.
- Klavis — Provides live environments for training agents with verifiable reward data.
- BellRun — Influence an autonomous AI intraday stock trader in a simulated rodeo market game.
- NPCTalk — Adds real-time cloned NPC voice servers and AI agent skills for game mods.
- MuscleMemory — It cuts input tokens, tool calls, and execution time while maintaining task success.
- Beeuni — Agents retain context and improve skills over time through a controlled review process.
All Pub Street Fighter AI alternatives