ITR: Self-Organizing Collective Intelligence for Adaptive Problem-Solving

University Of Maryland, College Park, College Park MD

Investigators

Abstract

Self-organizing particle systems consist of numerous autonomous "particles" (individuals) that move through space in a coordinated fashion, guided by influences that they exert on one another. Past computational methods based on such particle systems have primarily been used to produce striking computer animation effects, such as movement of a flock of birds or a school of fish, and to control movements of experimental robot teams. The goal of this project is to extend the methods currently used in particle systems to produce a goal-directed and adaptive collective intelligence. The individual particles have a limited capacity memory, prioritized goals, a repertoire of sequential movement patterns, and the ability to learn. The specific objectives are to create benchmark test problems for evaluating particle systems that have been extended in this fashion, to implement a goal-driven control mechanism for individual particles, to give particles the ability to learn during problem-solving, and to apply evolutionary computation approaches to create alternate control mechanisms. The methods used in this research include computer simulation, machine learning and evolutionary computation. The primary expected result is the creation of a multi-agent team that will function as an effective problem-solving system. In addition to providing insight into the fundamental computational principles underlying this approach to machine intelligence, the results from this research will have substantial practical value in developing natural scene computer animation, guiding the design of multi-robot teams, and suggesting ideas to biologists studying self-organizing processes in nature.