Towards Flexible Coordination of Large Scale Multi-Agent Teams

Abstract

Summary. As a paradigm for coordinating cooperative agents in dynamic environments, teamwork has been shown to be capable of leading to flexible and robust behavior. However, when teamwork is applied to the problem of building teams with hundreds of members, its previously existing, fundamental limitations become apparent. In this paper, we address the limitations of existing models as they apply to very large agent teams. We develop algorithms aimed at flexible and efficient coordination, applying a decentralized social network topology for team organization and the abstract coordination behaviors of Team Oriented Plans (TOPs). From this basis, we present a model to organize a team into dynamically evolving subteams, in order to flexibly coordinate the team. Additionally, we put forward a novel approach to sharing information within large teams, which provides for targeted, efficient information delivery with a localized reasoning process model built on previously incoming information. We have developed domain-independent software proxies, with which we demonstrate teams of an order of magnitude larger than those previously discussed in known published work. We implement the results of our approach, demonstrating its ability to handle the challenges of coordinating large agent teams.

Information searching and sharing in large-scale dynamic networks

“…(c) It supports the acquisition and exploitation of different types of locally available information for the 'tuning' process. (d) It extends the tokenbased method for information sharing (as it was originally proposed in [12,13]) in two respects: First, to deal with categories of information represented by means of ontology concepts and not with specific pieces of information, and second, to guide agents to advertise information that is semantically similar to the information requested, by using a semantic similarity measure between information categories. Therefore, it paves the way for the use of token-based methods for semantic peer-to-peer systems.…”

“…The proposed method requires a high volume of computations that authors aim to reduce by restricting its application to static logical teams of associated agents. In accordance to this approach, in [12,13,14], information sharing is considered only for static networks and self-tuning of networks is not demonstrated. As it will be shown in section 5, our experiments show that although these approaches can handle information sharing in dynamic networks, they require a larger amount of messages in comparison to the approach proposed here and can not tune the network for efficient information sharing.…”

“…one agent's actions affect the observations and the state of an other) [3], is hard. This fact, has resulted to efforts that either require agents to have a global view of the system [15], to heuristics [4], to precomputation of agents' information needs and information provision capabilities for proactive communication [17], to localized reasoning processes built on incoming information [12,13,14], and to mathematical frameworks for coordination whose optimal policies can be approximated [11] for small (sub-) networks of associated agents.…”

A multiagent framework for coordinated parallel problem solving

“…In order to achieve the joint goal, each team member maintains an explicit model of other agents and team activities at any moment. As [27] points out, maintaining models of all the co-participants introduces a significant amount of communication overhead, particularly when the number of team members are high. Therefore, methods limiting the communication overhead in coordination of the cooperating agents are needed.…”

“…This approach has demonstrated results in various areas such as speech understanding [12], visual shape recognition [9], vehicle monitoring [17], and medical diagnosis [26]. Recent progress in the multi-agent systems field led to the revisit of cooperative problem solving issue, this time from an agent-oriented perspective emphasizing the methods to coordinate the concurrent and autonomous behaviours of the collaborators [10,14,27,29].…”