On the Effects of Omitting Information Exchange between Autonomous Resource Management Agents

Abstract

We study the problem of information exchange between coordinated autonomous resource management agents. We limit information to that which can be directly observed by each agent. While exchanging all relevant observables leads to near optimal management, leaving out information leads to ”hidden variable” problems that affect only part of the overall behavior. The patterns observed for hidden variables in simulation predict what will happen in a realistic situation when not all information is interchanged. Through simulation, we observe that leaving out information results in non-optimal behavior of the resource management model when resource needs are decreasing, although the partial information model performs very well when resource needs are increasing.