Pac-Man Evolution
Classic arcade rules with evolving ghost intelligence.
Pac-Man Evolution is a C++17 game built with the CRM64Pro GDK, combining procedural maze generation with neural-network ghost AI trained through genetic evolution.
Overview
Pac-Man Evolution is a modern recreation of the classic arcade game featuring procedurally generated mazes and intelligent ghost AI powered by trained neural networks. It showcases advanced AI game development techniques by combining genetic algorithms with neural networks to create adaptive and cooperative ghost behaviors.
The game offers two distinct gameplay experiences: Standard mode with classic ghost behavior rules, and Evolution mode featuring neural network-driven ghosts trained over 400 generations for enhanced challenge and strategy.
Pac-Man is a trademark of Bandai Namco Entertainment.
This is an independent game inspired by the original concept and is not affiliated with or endorsed by Bandai Namco.
Screenshots
Features
Game modes
- Standard: Classic ghost behavior rules faithful to the original arcade experience.
- Evolution: Neural network-trained ghosts use advanced strategies, cooperative hunting patterns and emergent behaviors developed over 400 generations of genetic evolution.
- Workbench: Experimental developer mode for creating, testing and evolving custom ghost behaviors.
Neural network AI
- 4 trained ghost agents with distinct personalities and hunting strategies.
- Genetic algorithm training evolved over 400 generations to optimize ghost performance and cooperation.
- Ghosts learn to work together, using coordinated tactics to trap the player.
- Custom ghost brains are stored as XML files that can be exported, shared and modified.
- Integrated training system for evolving new behaviors with custom neural network architectures.
Procedural generation
- Every game features a unique, balanced maze layout.
- Maze algorithm adapted from Shaun LeBron’s JavaScript implementation.
- Deterministic generation keeps mazes fair and playable with balanced pellet distribution.
- No repeated layouts, keeping each run fresh and challenging.
Classic elements
- High score tables for each game mode.
- Power pellets to turn the tables on pursuers.
- Faithful controls, scoring system and gameplay feel.
Technical specifications
- Modern C++17 codebase built on the CRM64Pro GDK framework.
- Complete C++ source code available under the zlib license.
- Hardware-accelerated rendering for smooth 60 FPS performance at 1024×768 resolution.
- Audio system supporting high quality playback.
Operating systems supported
Cross-platform
- Windows 7, 10 and 11.
- Linux with kernel 4.18 or later.
- macOS 10.13 or later on x64 and arm64.
Development and contributions
The game is functionally complete and features a sophisticated AI training system, but is awaiting final graphics, music and sound effects. Contributions from artists, musicians and sound designers are welcome to help complete the project.
Contributors can use EditorC64 to enhance game resources.
- Graphics: enhanced sprite artwork, maze tiles, UI design and animations.
- Music: background tracks for menu and gameplay.
- Sound effects: pellet collection, power-ups, ghost interactions and menu feedback.
For developers and AI enthusiasts
- Access the workbench mode by compiling from source.
- Train custom ghost behaviors using the integrated neural network system.
- Export and share ghost brains as XML files with the community.
- Experiment with genetic algorithm parameters and neural network architectures.
- Analyze training performance and evolution progress in real time.
Download
Latest version: v1.0.2 (18-December-2025)
ChangeLog | Roadmap | zlib license
- Source code
- Source code and make files for VS2022+ and GNU C/C++.
- Previous releases
- Get access to all releases repository.


