In multi-robot systems, a large number of individual robots work together to perform a task. Large swarms of robots have the potential to be useful for types of tasks for which traditional robots are not well suited, such as large area surveillance, distributed contaminant cleanup, and in-vitro medical applications. Although a variety of possible robots could be used in such swarms, the large numbers involved usually force each robot to be small and simple. This dissertation focuses on robots with very simple hardware capabilities. In this work, the robots are not assumed to have global position information, global communication, odometry, or a central leader, and their only communication capabilities are simple local neighbor-to-neighbor communication. To overcome their hardware limitations and leverage their large numbers, the robots must work together.

This dissertation is focused on foraging as a multi-robot task. Robots start at a home location, explore the world in search of their target, and return the target incrementally to the home. It is difficult in swarm robotic systems because of the lack of global localization, communication, and odometry, which make it impossible for the robots to acquire or build maps, for example. Once a robot loses contact with the other robots, it is effectively lost and has no way (other than random movement) to return home.

This thesis begins by developing three distributed foraging algorithms for robot swarms and analyzing them in simulation. The performance of each algorithm is strongly affected by the environment in which the swarm operates, specifically the distance separating the home from the target. To allow the swarm to forage in different environments, a method is developed by which the swarm as a whole can choose its algorithm based on the home-target separation it encounters. Finally, I consider the feasibility of running these algorithms on physical robots. Adding accurate sensors and actuators to robots increases their cost, making it desirable for the robots to be as simple as possible while still being able to execute the algorithms. I directly measure the effect of sensor and actuator quality on swarm performance.