Publications

Abstract not provided.

Abstract not provided.

Abstract not provided.

This report describes a workshop on self-healing infrastructures conducted jointly by Sandia National Laboratories, Infrastructure & Information Division, and the Massachusetts Institute of Technology, Engineering Systems Division. The workshop was held in summer, 2002 and funded under Laboratory-Directed Research and Development (LDRD) No.5 1540. The purpose of the workshop was to obtain a working definition of a self-healing infrastructure, explore concepts for self-healing infrastructures systems, and to propose engineering studies that would lay the foundation for the realization of such systems. The workshop produced a number of useful working documents that clarified the concept of self-healing applied to large-scale system-of-systems exemplified by the US National Critical Infrastructure. The workshop eventually resulted in a joint proposal to the National Science Foundation and a continuing collaboration on intelligent agent based approaches to coordination of infrastructure systems in a self-healing regime.

Abstract not provided.

This report describes the results of research and development in the area of communication among disparate species of software agents. The two primary elements of the work are the formation of ontologies for use by software agents and the means by which software agents are instructed to carry out complex tasks that require interaction with other agents. This work was grounded in the areas of commercial transport and cybersecurity.

Intelligent agents and multi-agent systems promise to take information management for real-time control of the power grid to a new level. This report presents our concept for intelligent agents to mediate and coordinate communications between Control Areas and Security Coordinators for real-time control of the power grid. An appendix describes the organizations and publications that deal with agent technologies.

There are currently two proposed standards for agent communication languages, namely, KQML (Finin, Lobrou, and Mayfield 1994) and the FIPA ACL. Neither standard has yet achieved primacy, and neither has been evaluated extensively in an open environment such as the Internet. It seems prudent therefore to design a general-purpose agent communications facility for new agent architectures that is flexible yet provides an architecture that accepts many different specializations. In this paper we exhibit the salient features of an agent communications architecture based on distributed metaobjects. This architecture captures design commitments at a metaobject level, leaving the base-level design and implementation up to the agent developer. The scope of the metamodel is broad enough to accommodate many different communication protocols, interaction protocols, and knowledge sharing regimes through extensions to the metaobject framework. We conclude that with a powerful distributed object substrate that supports metaobject communications, a general framework can be developed that will effectively enable different approaches to agent communications in the same agent system. We have implemented a KQML-based communications protocol and have several special-purpose interaction protocols under development.

Our vision of the future of information systems is one that includes engineered collectives of software agents which are situated in an environment over years and which increasingly improve the performance of the overall system of which they are a part. At a minimum, the movement of agent and multi-agent technology into National Security applications, including their use in information assurance, is apparent today. The use of deliberative, autonomous agents in high-consequence/high-security applications will require a commensurate level of protection and confidence in the predictability of system-level behavior. At Sandia National Laboratories, we have defined and are addressing a research agenda that integrates the surety (safety, security, and reliability) into agent-based systems at a deep level. Surety is addressed at multiple levels: The integrity of individual agents must be protected by addressing potential failure modes and vulnerabilities to malevolent threats. Providing for the surety of the collective requires attention to communications surety issues and mechanisms for identifying and working with trusted collaborators. At the highest level, using agent-based collectives within a large-scale distributed system requires the development of principled design methods to deliver the desired emergent performance or surety characteristics. This position paper will outline the research directions underway at Sandia, will discuss relevant work being performed elsewhere, and will report progress to date toward assurance in agent-based systems.

Moving commercial cargo across the US-Mexico border is currently a complex, paper-based, error-prone process that incurs expensive inspections and delays at several ports of entry in the Southwestern US. Improved information handling will dramatically reduce border dwell time, variation in delivery time, and inventories, and will give better control of the shipment process. The Border Trade Facilitation System (BTFS) is an agent-based collaborative work environment that assists geographically distributed commercial and government users with transshipment of goods across the US-Mexico border. Software agents mediate the creation, validation and secure sharing of shipment information and regulatory documentation over the Internet, using the World Wide Web to interface with human actors. Agents are organized into Agencies. Each agency represents a commercial or government agency. Agents perform four specific functions on behalf of their user organizations: (1) agents with domain knowledge elicit commercial and regulatory information from human specialists through forms presented via web browsers; (2) agents mediate information from forms with diverse otologies, copying invariant data from one form to another thereby eliminating the need for duplicate data entry; (3) cohorts of distributed agents coordinate the work flow among the various information providers and they monitor overall progress of the documentation and the location of the shipment to ensure that all regulatory requirements are met prior to arrival at the border; (4) agents provide status information to human actors and attempt to influence them when problems are predicted.

One important application of mobile robots is searching a geographical region to locate the origin of a specific sensible phenomenon. Mapping mine fields, extraterrestrial and undersea exploration, the location of chemical and biological weapons, and the location of explosive devices are just a few potential applications. Teams of robotic bloodhounds have a simple common goal; to converge on the location of the source phenomenon, confirm its intensity, and to remain aggregated around it until directed to take some other action. In cases where human intervention through teleoperation is not possible, the robot team must be deployed in a territory without supervision, requiring an autonomous decentralized coordination strategy. This paper presents the alpha beta coordination strategy, a family of collective search algorithms that are based on dynamic partitioning of the robotic team into two complementary social roles according to a sensor based status measure. Robots in the alpha role are risk takers, motivated to improve their status by exploring new regions of the search space. Robots in the beta role are motivated to improve but are conservative, and tend to remain aggregated and stationary until the alpha robots have identified better regions of the search space. Roles are determined dynamically by each member of the team based on the status of the individual robot relative to the current state of the collective. Partitioning the robot team into alpha and beta roles results in a balance between exploration and exploitation, and can yield collective energy savings and improved resistance to sensor noise and defectors. Alpha robots waste energy exploring new territory, and are more sensitive to the effects of ambient noise and to defectors reporting inflated status. Beta robots conserve energy by moving in a direct path to regions of confirmed high status.