Talking helps: evolving communicating agents for the predator-prey pursuit problem

Area of Science:

• Artificial Intelligence
• Computational Biology
• Evolutionary Computation

Background:

• Multi-agent systems often require effective communication strategies for optimal performance.
• Simultaneous communication models and predator-prey pursuit problems are key areas in AI research.

Purpose of the Study:

• To evolve multi-agent communication languages for predator agents using a genetic algorithm.
• To analyze the impact of language size on predator performance and system behavior.
• To develop an efficient method for evolving complex communication systems.

Main Methods:

• Utilized a genetic algorithm to evolve predator languages in a predator-prey pursuit simulation.
• Modeled the communicating multi-agent system as a Mealy finite state machine.
• Introduced an incremental language size increase for efficient evolution.

Main Results:

• Evolved communication languages significantly improved predator performance.
• Increasing language size further enhanced predator effectiveness and system capabilities.
• The incremental approach reduced evolution time and resulted in smaller Mealy machines.
• Evolved predators outperformed previous state-of-the-art in similar prey scenarios.

Conclusions:

• Communication is crucial for enhancing multi-agent system performance in complex tasks.
• Language size is a critical factor influencing emergent behavior and task success.
• An incremental evolution strategy offers an efficient pathway to developing sophisticated agent communication.