ArtOfWar：融合群体智能与遗传算法的RTS游戏AI实验平台

ArtOfWar is an open-source RTS game inspired by Age of Empires, developed using the Urho3D engine. It integrates swarm intelligence (herd algorithms, crowd dynamics), genetic algorithms, and neural networks to provide a unique experimental environment for AI strategy research. This platform serves as a testbed for multi-agent AI studies, combining classic RTS gameplay with modern AI technologies.

即时战略游戏（RTS）are important testbeds for AI research due to their complex decision spaces and multi-agent collaboration. ArtOfWar延续 this tradition, inspired by Age of Empires. It uses the Urho3D engine (lightweight, cross-platform C++ engine) for底层 control and customization, which is crucial for integrating custom AI algorithms (instead of Unity/Unreal for more flexibility).

ArtOfWar's tech stack combines multiple AI subfields:

Herd Algorithm

Simulates large-scale unit movement with three core rules: Separation (avoid crowding), Alignment (same direction as neighbors), Cohesion (move to average position of neighbors) — solving unit stacking and path congestion.

Crowd Dynamics

Advanced swarm behavior for large units in complex environments: flow field path planning, density-aware movement, emergency obstacle avoidance.

Genetic Algorithms

Evolution engine for optimizing tactics (unit grouping, attack timing), economic decisions (resource balance), base construction (layout). Discovers emergent strategies beyond human design.

Neural Networks

Provides learning/adaptation: Value Network (state evaluation), Strategy Network (action probability), Prediction Model (opponent action预判). Combined with genetic algorithms for robust, adaptive systems.

Main language: C++ (for efficiency). Project structure:

• game/: core logic
• tests/: unit/integration tests
• .clang-format: code quality
• artofwar.sln: Visual Studio solution (Windows support). Version control via Git, with Issues tracking and Wiki docs for community maintenance.

ArtOfWar is more than a game — it's an AI research platform:

• Swarm Intelligence Research: Test herd algorithm variants in complex RTS scenarios.
• Evolutionary Algorithm Evaluation: Visualize genetic algorithm effects via battle results.
• Reinforcement Learning Environment: Integrate neural networks for RL training with custom reward functions.
• Education: Open-source reference for students to learn AI integration in games.

Key challenges and fixes:

• Performance: C++/Urho3D choice, algorithm time complexity control, spatial partitioning, multi-threading.
• Algorithm Coordination: Layered architecture and clear interfaces to ensure swarm, GA, neural networks work together.
• Playability Balance: Difficulty levels, AI behavior randomization, or 'cheat' compensation to balance challenge and fun.

Potential next steps:

• Deep Learning Integration: Use PyTorch C++ API for complex networks (Transformer, graph neural networks).
• Multi-Agent RL: Implement MARL for self-play to improve AI strategies.
• Player Behavior Learning: Let AI adapt to human players' tactics.
• Procedural Level Generation: Use genetic algorithms for auto-generating balanced maps/tasks.

ArtOfWar is a tech-driven platform exploring the synergy of swarm intelligence, evolutionary computing, and neural networks. It benefits AI researchers (visual testbed), game developers (AI integration example), and learners (open resource).

As an open-source project: it offers transparency (code review), extensibility (community contributions), and education value. Current Issues list has 4 pending items, inviting developers to contribute and join the community.