README.md

metadata

title: Diamond CSGO AI Player
emoji: 🎮
colorFrom: green
colorTo: blue
sdk: docker
pinned: false
app_file: app.py

Diamond CSGO AI Player 🎮

A web-based demo of the Diamond AI agent playing Counter-Strike: Global Offensive using diffusion models and reinforcement learning.

Features

• Real-time Keyboard Input: Use standard WASD controls and other keys to interact
• AI Agent: Pre-trained agent using diffusion-based world models
• Web Interface: No installation required, play directly in your browser
• Live Visualization: See the AI's perspective and actions in real-time

Controls

Movement

• W - Move Forward
• A - Move Left
• S - Move Back
• D - Move Right
• Space - Jump
• Ctrl - Crouch
• Shift - Walk

Actions

• 1, 2, 3 - Switch Weapons
• R - Reload
• Arrow Keys - Camera Movement
• Left/Right Click - Primary/Secondary Fire

Game Controls

• M - Switch between Human/AI control
• Enter - Reset Environment

How to Play

1. Click on the game canvas to focus it
2. Use keyboard controls to play
3. The AI agent will respond to your inputs in real-time
4. Switch to AI mode to watch the agent play autonomously

Technical Details

This demo uses:

• FastAPI + WebSocket for real-time communication
• PyTorch for AI model inference
• Diffusion Models for next-frame prediction
• World Model Environment for simulation

The agent was trained using the Diamond framework, which combines:

• Diffusion-based world models
• Actor-critic reinforcement learning
• Multi-step planning and imagination

Model Information

The AI agent uses several neural networks:

• Denoiser: Diffusion model for generating next observations
• Upsampler: High-resolution image generation
• Reward/End Model: Predicting game outcomes
• Actor-Critic: Action selection and value estimation

Citation

This work is based on the Diamond framework. If you use this code, please cite:

@article{diamond2024,
  title={Diamond: Diffusion for World Modeling},
  author={[Authors]},
  journal={[Journal]},
  year={2024}
}

License

See LICENSE file for details.