An open-source project enabling an AI model to play the popular mobile game Honor of Kings (王者荣耀). It targets AI researchers and developers interested in game AI, providing a platform for learning and experimentation with AI agents in a complex gaming environment. The project aims to advance AI capabilities in real-time strategy games, distinct from cheating applications.

How It Works

The project involves training AI models, likely using techniques such as imitation learning or reinforcement learning, leveraging ONNX models as a starting point. Gameplay logic is implemented through Python scripts that interact with the game environment, potentially via screen analysis and input simulation. Key aspects include configuring input mappings and screen coordinates for accurate game control, with a training script (train.py) generating playable models (.pt files).

Quick Start & Requirements

• Installation: Requires Anaconda for environment management. Create and activate a conda environment (conda create --name wzry_ai python=3.10, conda activate wzry_ai), then install dependencies using pip install -r requirements.txt.
• Prerequisites: Python 3.10, Anaconda, PyTorch with CUDA 11.8 support (pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118), and ONNX Runtime GPU (specific installation for CUDA 11 or 12). A detailed installation tutorial is available via a Bilibili video.
• Setup: Environment setup involves multiple steps and specific CUDA/PyTorch versions, potentially requiring troubleshooting for components like zlibwapi.dll.
• Links: Anaconda Download: https://www.anaconda.com/download, Bilibili Tutorial: https://www.bilibili.com/video/BV1ZXYuePEUG/

Highlighted Details

• Focuses on AI learning and research, explicitly prohibiting cheating applications.
• Supports training custom models from provided ONNX files.
• Includes utilities (showposition.py) to help map game interface coordinates for AI control.

Maintenance & Community

• Active development with Phase 1 complete and Phase 2 underway.
• Community support available via QQ groups (two groups listed, one full).
• Installation guidance provided via Bilibili video.

Licensing & Compatibility

• License type is not specified in the provided README.
• Compatibility is primarily geared towards Windows environments due to specific DLL path references in setup instructions.

Limitations & Caveats

The project explicitly states its purpose is for AI learning and strictly prohibits cheating, indicating it is not intended for competitive or unauthorized use. Setup requires specific CUDA versions and may involve complex environment configuration, potentially posing a barrier to entry. The README does not detail performance benchmarks or specific AI algorithms used.