Development And Analysis of Biologically Inspired Communication Algorithms for Artificial Agents

Abstract

The capabilities of robots have increased to the point where it is more effective and economical to field a large group of less expensive robots than to field a single excessively costly robot. A consistent and effective system for managing large amounts of robots in a swarm must be developed to expand this capability for regular application. Inspiration for these expansive communications systems can be found in the biological world. Many species, whether birds, fish, or insects, have demonstrated powerful swarming capabilities across a wide variety of tasks. Models from the biological literature have demonstrated properties and capabilities that make them ideal for swarm robotics.

This thesis identifies three relevant swarm communications models and develops algorithms to allow these models to interact in a robotic swarm environment. Using a simulation, the three models are evaluated over a series of tasks chosen to approximate swarm robotics tasks. The eight tasks included searching for goals, avoiding adversaries, and controlling the position and density of the swarm. The capabilities of each model - closer or farther sensing capabilities or unique target selection - provided varying performances on each of the tasks. The results indicated that specific swarm communications models need to be selected for each task to achieve optimal results.