| Swarm engineering is the natural evolution of the use of swarm-based techniques in the accomplishment of high level tasks using a number of simple robots. In this approach, one seeks not to generate a class of behaviors designed to accomplish a given global goal, as is the approach typically found in mainstream robotics.; This approach is investigated in terms of three specific practical problems. First we apply swarm engineering to plume tracking, utilizing both real and simulated experiments.; The second problem is position-independent clustering, in which a cluster of objects is constructed from many initial clusters, without predetermination of the final cluster's location.; The third and last problem is the traveling salesman problem (TSP). Marco Dorigo (see, for instance, Dorigo 1996) generated much interest in the use of “ants” on the problem. We theoretically demonstrate that Dorigo's ants satisfy a swarm engineering condition generated for this problem.; This thesis makes a number of contributions to the literature. First, we study two systems that take advantage of the embodiment and interaction dynamics of the robot to accomplish the single robot goal. We then extend this work to answer two questions: (1) Is it possible to apply swarm engineering to plume tracking? (2) How can one apply swarm engineering to plume tracking? We demonstrate that is indeed possible to extend one of the two previous plume tracking systems to a swarm. We demonstrate that the swarm is capable of successfully tracking plumes that are significantly more diffuse than those capable of being tracked by individual agents.; Next, we derive a formal condition which must be satisfied in order for a system made up of many clusters and robots to become a system of one cluster and many robots, if the robots do not have global information.; Our third contribution is a formalism for ant-based TSP algorithms. This allows us to understand why Dorigo's algorithms are relatively unstable.; Perhaps the most important contribution here is an alternative approach to the design of swarm systems. This allows us to explore the three problems of interest here by first determining a condition that allows the completion of the task, and then allows the generation of behaviors satisfying the minimal condition. (Abstract shortened by UMI.)... |
