Team Member Reallocation via Tree Pruning

This paper considers the task reallocation problem, where agents are to be extracted from a coordinated group of agents in order to perform a new task. The interaction between the team members and the cost associated with this interaction are represented by a weighted graph. Consider a group of robots organized in a formation, the graph is the monitoring graph which represents the sensorial capabilities of the robots, i.e., which robot can sense the other and at what cost. Following this example, the team member reallocation problem this paper deals with is the extraction of robots from the group in order to acquire a new target, while minimizing the cost of the interaction of the remaining group. In general, the method proposed here shifts the utility from the team member itself to the interaction between the members, and calculates the reallocation according to this interaction utility. We found that this can be done optimally by a deterministic polynomial time algorithm under several constraints, the first constraint is that . We describe several other domains in which this method is applicable. Introduction This article discusses a team of agents engaged in a cooperative behavior (Kraus 1997). Specifically, we consider the problem of choosing team members in order to assign them to a new task. We assume that all members are capable of performing the new task, and the cost of the new task does not depend on the identity of the agents chosen to perform it. Therefore this paper concentrates on the problem of choosing agents such that the performance of the existing task, performed by the remaining group members, will be as efficient as possible. The measure according to which the reallocation is done is based on the interaction between team members. Therefore the efficiency in this context refers to the cost of interaction, i.e., as the team spends less resources on interaction, the execution of the task is more efficient. The set of team members and the interaction between them is represented by a weighted directed graph, where the vertices represent the members, and the edges represent the inGal Kaminka is also affiliated with CMU and Sarit Kraus is also affiliated with UMIACS. This research was supported in part by NSF grant # IIS