Catálogo de publicaciones - libros

In this paper, we describe technologies for an agent server capable of hosting millions of agents. The agent server needs a thread management mechanism, a memory management mechanism, and a recovery management mechanism. We have developed a framework and agent execution environment named Caribbean. First, we describe the programming model of Caribbean. Following the description, we explain technologies for managing millions of agents. Some application scenarios of real commercial systems using the technology are also introduced. We describe what we learned from the development of the real applications.

The Intelligent Agent Grid Environment is a scalable, live, sustainable and intelligent networking environment where humans, agents, machines and nature can harmoniously co-exist, work and evolve. It automatically collects useful information from nature and society according to requirements, transforms it into resources in the environment, and then after intelligent processing, affects nature and society through machines. According to the regulations and principles of the environment, people, agents and resources can intelligently cooperate with each other to accomplish tasks, generate knowledge and solve problems by actively participating in versatile flow cycles in the environment through roles and machines. The Environment is the unity of the natural material world, virtual world and cognitive world. Various types of attraction in the environment drive the flows. The rules of flows guide the development and management of the environment.

Although distributed computing is necessary to execute massively multi-agent applications, the distribution of agents is challenging especially when the communication patterns among agents are continuously changing. This paper proposes two adaptive agent allocation mechanisms for massively multi-agent applications: one mechanism aims at minimizing agent communication cost, while the other mechanism attempts to prevent overloaded computer nodes from negatively affecting overall performance. We synthesize these two mechanisms in a multi-agent framework called (). In AAA, each agent platform monitors the workload of its computer node and the communication patterns of agents executing on it. An agent platform periodically reallocates agents according to their communication localities. When an agent platform is overloaded, the platform migrates a set of agents, which have more intra-group communication than inter-group or inter-node communication, to a lightly loaded agent platform. These adaptive agent allocation mechanisms are developed as fully distributed algorithms, and they move the selected agents as a group. In order to evaluate these mechanisms, preliminary experimental results with large-scale micro UAV (Unmanned Aerial Vehicle) simulations are described.

Scalable coordination is a key challenge in deploying massively multi-agent systems. Resource usage is one part of agent behavior which naturally lends itself to abstraction. CyberOrgs is a model for hierarchical coordination of resource usage by multi-agent applications in a network of peer-owned resources. Programming constructs based on CyberOrgs allow resource trade and control reification while maintaining a separation between functional and resource concerns. An operational semantics of CyberOrgs is presented. Expressive power of programming constructs based on CyberOrgs is illustrated with examples.

Hierarchical control presents challenges in scalability. However, some types of resource coordination are amenable to efficient implementation using CyberOrgs. Hierarchical control of processor time, for instance, can be implemented scalably by efficiently flattening the hierarchical schedule on the fly. Experimental results demonstrate scalability of the technique. Generalizations of this solution for hierarchical control of processor, network and other computational resources in a distributed system are discussed.

In order to construct and deploy massively multiagent systems, we must address one of the fundamental issues of distributed systems, the possibility of partial failures. In this paper, we discuss the issues and propose an approach for fault-tolerance of massively multiagent systems. The starting idea is the application of replication strategies to agents. As criticality of agents may evolve during the course of computation and problem solving, and as resources are bounded, we need to dynamically and automatically adapt the number of replicas of agents, in order to maximize their reliability and availability. We will describe our approach and related mechanisms for evaluating the criticality of a given agent and how to parameterize it (e.g., number of replicas). We also will report on experiments conducted with our prototype architecture (named DarX).

Interactive multi-agent system improves reusability of agents by separating application design from agent design. However, it remains difficult for application designers (usually non-computer professionals) to script massive multi-agents. This is especially true for scripting hundreds of s (Non Player Characters, agents) hosted by hostile, dynamic and complex 3D (three-dimensional) environments in military simulation. Out of perspective of , namely virtual spaces, we introduce to facilitate . Three advantages are obtained by using : 1) A hybrid approach of integrating qualitative and quantitative spatial reasoning is achieved so that application designers can specify arguments of s (scenario primitives) . 2)Primitive actions of agents are abstracted hierarchically so that application designers can scenarios for units of agents. 3) Better intent communication between humans and agents is realized so that application users can control agents easily in real time.

When agents coordinate according to the principles of teamwork they can flexibly, robustly and reliably achieve complex goals in complex, dynamic and even hostile environments. An emerging standard for building such teams is via the use of , which encapsulate domain independent teamwork algorithms in a software module that works closely with a domain specific person, agent or robot and other proxies to create a team. Succesful, previous generations of proxies and teamwork algorithms were limited to small teams because of their reliance on accurate models of the team and task state. By developing new algorithms that rely on probabilistic models we have been able to build teams that are orders of magnitude larger than before. However, key challenges remain before such teams can be deployed in real-world environments, including the need for languages to specify plans for such teams and ways of modeling and predicting team performance in new domains.

We present a fully distributed algorithm for coalition formation among autonomous agents. The algorithm is based on two main ideas. One is a distributed computation of maximal cliques (of bounded sizes) in the underlying graph that captures the interconnection communication topology of the agents. Hence, given the current configuration of the agents, the coalitions that are formed are characterized by a high degree of connectivity, and therefore a high fault tolerance with respect to the subsequent node and/or link failures. The second idea is that each agent chooses its most preferable coalition based on how highly the agent values each such coalition in terms of the coalition members’ combined resources or capabilities. Coalitions with sufficient resources for fulfilling are preferable to the coalitions with resources that suffice only for completing less valuable tasks. We envision variants of our distributed algorithm presented herein to prove themselves useful coordination subroutines in many massively multi-agent system applications where the agents may repeatedly need to form temporary groups or coalitions of modest sizes in an efficient, online and fully distributed manner.

As the scale and scope of multi-agent systems grow, it becomes increasingly important to manage the manner in which the participants interact. The potential for bottlenecks, intractably large sets of coordination partners, and shared bounded resources can make individual and high-level goals difficult to achieve. To address these problems, many large systems employ an additional layer of structuring, known as an organizational design, that assigns agents particular and different roles, responsibilities and peers. These additional constraints can allow agents to operate effectively within a large-scale system. In this paper, we will introduce a domain-independent organizational design representation capable of modeling and predicting the quantitative performance characteristics of agent organizations. This representation supports the selection of an appropriate design given a particular operational context. We will demonstrate how the language can be used to represent complex interactions, and show modeling techniques that can address the combinatorics of large-scale agent systems.

This paper describes the fundamentals of a research project which is being launched in the emerging field of as defined by the European Union’s 6th Research Program on Information Society. Massively multi-agent systems is the natural technique for implementing Ambient Intelligence. Adaptivity is one of the key features of ambient systems. Ensuring that the evolution of an ambient system is predictable and desirable is a challenging open design issue. We propose a user-driven approach to adaptation. We call it “Adaptive Modeling” because it relies on the architectural style known as Adaptive Object-Models. This provides us with a design method and tool for agents to be used in this context. Systems built with this method allow non-programmer domain experts to locally modify the structure and behavior of agents at runtime, and thus obtain system-level adaptation. Expert-driven adaptation should ensure the appropriateness of the system’s behavior with respect to its requirements. We illustrate our method with an existing multi-agent system. Work is under way for extending it with other features, notably fault-tolerance, as well as “agent-driven adaptation” by replacing expert users with monitoring agents endowed with the same expertise.