Overview
Multi-agent simulations in game engines generate realistic human behaviors for synthetic training data. This project implements generalist agent training across Unreal, Unity, and Three.js environments using a hybrid reinforcement learning and machine learning approach.
Architecture
Individual agents operate under RL policy with in-engine inference. Language model integration enables action interpretation and planning, with real-time visualization of agent internal states within the simulation environment.
Applications
Urban and Transportation Planning
Agent-based modeling for transportation networks and urban development scenarios. Integrated with Unreal city samples project for large-scale urban environment simulation.
Supply Chain Logistics
Multi-agent coordination for logistics optimization and supply chain management scenarios.
Epidemiological Modeling
SEIR (Susceptible, Exposed, Infectious, Recovered) module combines disease spread modeling with agent-based simulation. Models how transportation patterns and urban mobility affect transmission dynamics in urban environments.
Civilization Simulations
Historical simulation framework modeling human civilization development. Implemented with Unity’s Happy Harvest 2D template for rapid prototyping.
Technical Details
- Multi-engine support: Unreal, Unity, Three.js
- Hybrid RL/ML agent architecture
- In-engine policy inference
- LLM-driven action planning
- Real-time agent state visualization