Providing Guaranteed Behaviors for Groups of Low-Capability Mobile Agents

Abstract

I consider the problem of designing algorithms for coordinated groups of low-capability, errorprone mobile agents. It is my thesis that some important collective behaviors of groups of lowcapability mobile agents can be guaranteed by the application of provably-correct distributed algorithms. This thesis is demonstrated by the provision of new algorithms that guarantee mobile agent behavioral properties in the face of noise and agent error, as well as proofs that characterize the requirements of particular tasks. The contributions of this dissertation include both algorithms for solving classes of tasks that are foundational to processes involving groups of mobile agents and proofs regarding the impact limited capabilities have on the ability of agents to perform various tasks. The capability proofs lie in two principle areas. The first set of proofs relate to the differences between stigmergic and broadcast communication; these proofs bound the time and number of additional agents required to emulate broadcast communication using stigmergy. The second set of proofs bound the capabilities needed to ensure that agents are able to locate one another in an unknown environment. In addition to these stand-alone proofs, there are also proofs accompanying each algorithm presented. The main classes of tasks solved relate to agents forming and maintaining a group. A family i Abstract ii of algorithms is provided, each of which guarantees rendezvous in bounded time for some class of agent capabilities. For agents with bounded-error knowledge of time and place these rendezvous algorithms are within a logarithmic term of being asymptotically optimal. A technique is presented for generating pair-wise cohesion constraints for various agent types; these constraints provide each agent with a set of allowable behaviors that provably keep the agents connected. vi multiple objectives and constraints into a single behavioral policy. My time as a masters' student at Brigham Young University was invaluable, being my first experience with research and a broad introduction in many projects. My advisor, Robert P. Burton, gave me the freedom to begin exploring AI formally within the context of my thesis on simulations for multidimensional time. Formal mentoring roles given me by Robert P. Burton and Sean Warnick and research projects supervised by Robert P. Burton, Michael Jones, and Sean Warnick in several topics were educational experiences and invaluable in helping me to develop the maturity to stick to a single topic for the years required to attain a Ph.D. Finally, I am deeply indebted to the members of my family, my local church congregation, my fellow graduate students, and my Psandbox compatriots for their friendship and encouragement throughout. I researched without them, but they kept me sane as I did so.