Development of swarm behavior in artificial learning agents that adapt to different foraging environments

Area of Science:

• Artificial intelligence
• Collective behavior modeling
• Computational physics

Background:

• Collective behavior and swarm formation are studied across biology and physics.
• Artificial learning agents offer a novel approach to modeling complex group dynamics.

Purpose of the Study:

• To model collective behavior using Projective Simulation.
• To investigate emergent motion patterns based on resource proximity.

Main Methods:

• Applied Projective Simulation to artificial learning agents.
• Utilized a reinforcement learning framework for agent decision-making.
• Analyzed agent trajectories in one-dimensional foraging scenarios.

Main Results:

• Strongly aligned swarms emerged when food resources were distant.
• Agents trained for distant resources exhibited Lévy walk-like trajectories.
• Agents trained for nearby resources showed predominantly Brownian trajectories.

Conclusions:

• Emergent collective motion is sensitive to foraging distance.
• Individual agent trajectories reflect the learned collective behavior and environmental pressures.
• Projective Simulation effectively models diverse swarm dynamics.