Using reservations for multiagent resource allocation with costly preemption

Abstract

We examine multiagent resource allocation to maximize global utility in settings with dynamic arrivals, proposing a distributed coordination framework that allows resource preemptions. We then introduce the concept of resource reservations to address expected future changes in the task allocation e nvironment; accounting for switching costs and considering irrevocable allocations both become desirable. We address these problems using dummy agents and validate our approach using simulations comparing utility with and without dummy agents. We discuss the value of this framework in the context of social computing, enabling improved interactions for teamwork and coordination in domains like health care for emergency response management.