Multi-Agent Pursuit-Evasion Game
Active
Overview
Two AI agents learn to play a game of chase—one tries to catch, the other tries to escape. They navigate around obstacles in an arena, improving their strategies through trial and error using reinforcement learning.
Objectives
- Train agents that can chase and evade intelligently
- Handle environments with varying obstacle layouts
- Transfer learned behaviors to physical robots
Methodology
Agents learn through self-play, gradually facing harder challenges as they improve. The system uses deep reinforcement learning with neural networks that adapt to different obstacle configurations.
Current Progress
- Training pipeline complete
- Agents successfully navigate obstacle environments
- Preparing for real robot deployment