Mario AI Base ๐ฎ๐ค๐
Welcome to Mario AI Base! ๐ This repository is your hub for an AI-powered Super Mario Bros. experience. An AI agent, trained with reinforcement learning, plays the game and streams its gameplay live to your browser. A Gradio-based updater app lets you manage daily notes from your LAN. Dive into retro gaming with a modern AI twist! ๐
Overview ๐
- Mario AI Streaming App: An AI (PPO from
stable-baselines3
) plays Super Mario Bros., streamed at ~30 FPS tohttp://<your-rig-ip>:5111
via MJPEG with Flask. Supports CUDA acceleration. - Python Updater App: Gradio-based tool to update GitHub notes daily, accessible at
http://<your-rig-ip>:2332
. - No ROM Included: Users must provide their own Super Mario Bros. ROM due to licensing. ๐น๏ธ
Features โจ
- AI-Driven Gameplay: Watch an AI navigate Mario levels! ๐ง
- Live Streaming: Real-time gameplay in your browser. ๐บ
- CUDA Acceleration: Faster AI inference on GPUs like RTX 3060 Ti. โก
- Daily Updates: Log progress with the Gradio app. ๐
- LAN Accessible: Stream and control from your network. ๐
- Customizable: Adjust AI model or add gameplay actions. ๐ฎ
Setup ๐ ๏ธ
Prerequisites ๐
- OS: Ubuntu 20.04+ (tested on "T-1000" with RTX 3060 Ti).
- Python: 3.8+ (
python3 --version
). - GPU (Optional): NVIDIA GPU (e.g., RTX 3060 Ti) for CUDA.
- NES ROM: Legal
Super Mario Bros.nes
(not provided) at/home/ubuntu/mario/
. - Git:
sudo apt install git
for the updater. - Dependencies: Install with
requirements.txt
.
Hardware Requirements ๐ป
- CPU: 2+ cores, 2 GHz (e.g., Intel i5). More for training.
- GPU: RTX 3060 Ti or similar for CUDA (recommended).
- RAM: 16 GB minimum (32 GB for training).
- Storage: 1 GB for dependencies, ROM, and models.
CUDA Setup for GPU Acceleration โก
Leverage your RTX 3060 Ti for AI inference and frame processing:
- Install NVIDIA Drivers:
sudo apt update sudo apt install nvidia-driver-550
Website: https://ghostai.pro/ Discord: https://discord.gg/9cnJNBQtHE
Update (March 13, 2025) ๐
- Added fresh enemy tactics for extra difficulty ๐
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API:
The model has no library tag.