GeMS: A multi-agent system using market-based control and generational scheduling for real-time distributed resource-allocation.

摘要

This thesis details the design of a unique and simple end-to-end solution for allocating tasks to heterogeneous resources, especially mobile, in a dynamic real-time environment called generational scheduling-based multi-agent system (GeMS reads as gems). The solution incorporates location information and travel times of resources. GeMS is a generic, fault-tolerant, scalable, and computationally light solution that eliminates oscillation, and minimizes cascading. It is a heuristic solution based on the "good enough quick enough" principle. The GeMS design is based on generational scheduling, and a multi-agent system that uses a market-based control for negotiations among agents. The implementation of GeMS is beyond the scope of this thesis.; GeMS emerged and evolved from a preliminary proof-of-concept solution called JavaSpaces(TM) based multi-agent system (SpaceMAS reads as space mass). We implemented and deployed SpaceMAS in a laboratory of five computers, several radio-controlled robotic vehicles, and a camera.