Evolving flocking in embodied agents based on local and global application of Reynolds' rules

Area of Science:

• Robotics
• Artificial Intelligence
• Collective Behavior

Background:

• Flocking is crucial for large-scale embodied agent systems like robot swarms.
• Evolving self-organized flocking behaviors using artificial techniques is challenging due to unknown conditions.

Purpose of the Study:

• To demonstrate how evolutionary techniques can synthesize flocking behaviors.
• To investigate optimal fitness function designs for evolving high-performing controllers in robot swarms.

Main Methods:

• Utilized evolutionary algorithms to evolve controllers for embodied agents.
• Designed fitness functions based on Reynolds' boids model rules (separation, cohesion, alignment).
• Modified fitness functions to incorporate global group considerations versus individual rule conformity.

Main Results:

• Embedding Reynolds' rules in fitness functions successfully evolved flocking behaviors.
• Individualistic rule adherence led to fragmented, local flocking.
• Global flocking emerged when fitness functions considered the entire agent group.
• Explicitly rewarding alignment was not necessary for successful flocking evolution.

Conclusions:

• Fitness function design is critical for evolving collective behaviors in agent systems.
• Global, cohesive movement in robot swarms can be achieved through evolutionary synthesis.
• Evolutionary techniques offer a pathway to creating sophisticated collective behaviors for embodied agents.