Agent Division and Fusion for Task Execution in Undependable Multiagent Systems

Abstract

In multiagent systems, agents with limited capacity often cooperate in order to accomplish various types of tasks. Due to the openness of multiagent systems, agents may run the risk of their cooperations to accomplish tasks because of some involved undependable agents. The state of the art for handling this problem concentrates on the phase of task allocation, which aims to allocate tasks to more dependable agents (task allocation-oriented). In fact, accomplishing a task in a multiagent system requires two phases: i) task allocation, and ii) task execution. Hence, this paper, on the contrary, focuses on managing the phase of task execution to guarantee the performance of task accomplishment (task execution-oriented). To reduce the performance loss caused by undependable agents, the proposed agent division and fusion mechanism enables agents to autonomously divide themselves into sub-agents for executing tasks with different risks (more resources will be assigned to the sub-agent with lower risk in task execution), then the sub-agents can also fuse together and make a re-division to fit the current task environments. This work is also expected to be able to complement the existing state of the art (task allocation-oriented) for guaranteeing the performance of task accomplishment in undependable multiagent systems from task execution-oriented perspective.