Trace an adaptive organizational policy for multi agent systems

Abstract

This thesis proposes an adaptive organizational policy, called TRACE (Task and Resource Allocation in a Computational Economy), for MAS that operate under time constraints and varying computational loads. We view the MAS as consisting of several problem-solving organizations where each organization is comprised of multiple agents that may be grouped into teams for specific problem solving. Problem solving requests arrive at each of these organizations. A request that arrives at an organization is solved cooperatively by agents within that organization and independently of the other organizations. As the rate of arrival of problem solving requests at each of these organizations varies with time, there could be a situation where some organizations may have surplus resources, while others have insufficient resources and thereby turn down problem solving requests. In order to minimize these lost requests, the allocation of resources (agents) to organizations is changed dynamically using the microeconomic approach. This reallocation of resources changes the number of agents in the organizations and their skills, and is intended to balance the demand for resources at each organization with its supply. Following a layered approach, we divide the problem of developing an adaptive organizational policy into two broad sub-problems viz. 1. Allocation of tasks to agents within an organization through the task allocation protocol (TAP). 2. Allocation of resources (agents) to each of these organizations using the resource allocation protocol (RAP).