Two Paradigms for the Design of Artificial Collectives

Summary. Artificial collec tives are sys tems composed of multiple autonom ous information or software agents, mobile robots, or nodes in a sensor or communication net work . In the future , such systems will be responsible for many important tasks, such as highway traffic co ntrol, disaster respon se, tox ic spill monitoring and cleanup, and explor ation of other plan­ ets. Because such sys tems will have to function in environments with unreliable communica­ tion channels, where agen ts are likely to fail, they will have to be reliable, sca lable, robu st, adaptable, and amenable to quantitative mathematical analysis. The last property is important because analysis is crucial to understandin g the issues of the desig n, co ntro l, ada ptability, and dynamics of collectiv e behavior. We describe two approac hes to distributed co ntrol of artificia l collectives and study them quantitatively. The first, biologically based contro l, relie s on local interactions among many simple age nts to create desirable collective beh avior. Th e sec ond ap­ proach allows collectives to maximize their world utility using market-based mechanisms. We present two applications- fora ging in a group of robots and resource allocat ion in dyn amic environment s-that use the se co ntrol paradigm s and perform an analysis of each problem.