CS2 Reinforcement Learning Agent

A reinforcement learning agent for Counter-Strike 2 that uses multiple training modes and computer vision to learn and play the game.

Directory Structure

cs2_rl_agent/
│
├── scripts/                  # Executable scripts
│   ├── training/             # Training script batch files
│   ├── testing/              # Testing script batch files  
│   ├── utils/                # Utility script batch files
│   └── deployment/           # Deployment script batch files
│
├── training/                 # Training Python scripts
│   ├── train_adaptive.py     # Adaptive agent training
│   ├── train_strategic.py    # Strategic agent training
│   ├── train_discovery.py    # Discovery mode training
│   ├── train.py              # General training script
│   └── ...
│
├── testing/                  # Testing Python scripts
│   ├── test_cs2_env.py       # CS2 environment tests
│   ├── test_config.py        # Configuration tests
│   ├── test_api.py           # API testing script
│   ├── test_vision_windows.py # Vision interface tests
│   └── ...
│
├── evaluation/               # Evaluation scripts
│   └── evaluate.py           # Model evaluation script
│
├── utils/                    # Utility Python scripts
│   ├── setup_gpu.py          # GPU setup utilities
│   ├── check_gpu.py          # GPU checking utilities
│   └── ...
│
├── config/                   # Configuration files
│   ├── adaptive_config.yaml  # Adaptive agent configuration
│   ├── strategic_config.yaml # Strategic agent configuration
│   └── ...
│
├── docs/                     # Documentation
│   ├── ANACONDA_SETUP.md     # Anaconda setup guide
│   ├── WINDOWS_SETUP.md      # Windows setup guide
│   └── ...
│
├── src/                      # Core source code
│   ├── agent/                # Agent implementations
│   ├── actions/              # Action system
│   ├── environment/          # Environment implementations
│   ├── utils/                # Core utilities
│   └── interface/            # Interface components
│
├── analysis_log/             # Analysis documentation
│
├── images/                   # Image files
│
├── models/                   # Trained models
│
├── data/                     # Training data
│
├── logs/                     # Log files
│
└── tensorboard/              # TensorBoard logs

Setup

1. Install Anaconda or Miniconda
2. Run the setup script: scripts/utils/setup_conda.bat
3. Install dependencies: pip install -r requirements.txt
4. Configure GPU (optional): scripts/utils/enable_gpu.bat

Training

To train the agent, use one of the following scripts:

• Adaptive Agent: scripts/training/train_adaptive.bat - Primary training script (recommended)

The adaptive agent serves as the central orchestrator that manages all specialized agent types:

• Discovery Mode
• Tutorial Mode
• Vision Mode
• Autonomous Mode
• Strategic Mode

This centralized approach eliminates the need for separate training pipelines while maintaining the full functionality of all specialized agent types.

Testing

The project includes comprehensive test scripts to verify functionality:

• Adaptive Orchestration Test: scripts/testing/test_adaptive_orchestration.bat - Tests the adaptive agent's ability to switch between different modes
• Vision System Test: scripts/testing/test_vision_windows.bat - Verifies the vision processing functionality
• Environment Test: scripts/testing/test_cs2_env.bat - Tests the game environment interface

The adaptive orchestration test is particularly valuable for verifying the streamlined architecture, as it confirms that:

1. The adaptive agent correctly initializes all specialized modes
2. Dynamic mode switching works based on performance metrics
3. Knowledge is transferred between different modes
4. The system handles errors gracefully

For quick testing, use:

scripts\testing\test_adaptive_orchestration.bat

For more comprehensive testing with the real environment, use:

scripts\testing\test_adaptive_orchestration.bat --real-env 300

Deployment

For deployment, use one of the following scripts:

• All-in-One Setup and Train: scripts/deployment/all_in_one_setup_and_train.bat - Complete setup and training
• Run Adaptive Agent: scripts/deployment/run_adaptive_agent.bat - Focused deployment script

These streamlined deployment scripts provide a simplified interface while maintaining full functionality through the adaptive agent's orchestration capabilities.

Documentation

For a complete understanding of the architecture, refer to these key documents:

• Executive Summary: analysis_log/architecture/adaptive_architecture_executive_summary.md - High-level overview of the streamlined architecture
• Architecture Details: analysis_log/architecture/adaptive_orchestration.md - Detailed explanation of the adaptive agent orchestration
• Visual Diagram: analysis_log/visualization/adaptive_orchestration_diagram.md - Visual representation of the architecture
• Testing Documentation: analysis_log/testing/adaptive_orchestration_testing.md - Details of the testing methodology

For detailed documentation, see the following:

• Anaconda Setup Guide
• Windows Setup Guide
• All-in-One Guide
• Autonomous Agent Guide
• Auto Detection Guide

For code analysis and architecture documentation, see the analysis_log directory.

License

This project is licensed under the MIT License - see the LICENSE file for details.