In this paper we develop a new model of task allocation in distributed and cooperative resource-bounded agents using a theoretic-decision approach and their effect on the responsiveness of the system. Two architectures for task allocation and their level of deliberation are discussed. In both architectures the following holds: (1) Agents have limited resources and estimated distributions over resource execution tasks and (2) Agents create new tasks that they send to a central controller to distribute among them. The main difference between the two architectures resides in the place where the allocation decision-making process is performed. In the first architecture, we assume that the central controller builds an optimal and global decision on task allocation using a dynamic programming model and an estimated distribution over resources. In the second architecture, we assume that each agent builds a locally optimal decision and the central controller coordinates these distributed locally optimal decisions. In both architectures, we formulate the standard problem of task allocation as a Markov Decision Process (MDP). The states of the MDP represent the current state of the allocation in terms of tasks allocated to each agent and available resources. It is well-known that such approaches have a high-level of deliberation that can affect their efficiency in dynamic situations. We then discuss the effect of the two architectures on the balance between the deliberative and reactive behavior of the system.
展开▼
