cast agents: network-centric fires unleashed

12
Distribution Statement A 1 Distribution Statement A: Cleared for Public Release; distribution is unlimited. SPAWAR Security & Policy Review control number: SR-2001-098. CAST Agents: Network-Centric Fires Unleashed Martin O. Hofmann, Daria Chacón, Gerard Mayer, Kenneth R. Whitebread Lockheed Martin Advanced Technology Laboratories Camden, NJ 08102 {mhofmann, dchacon, gmayer, kwhitebr}@atl.lmco.com James Hendler Chief Scientist, DARPA/ISO Arlington, VA 22203 [email protected] Abstract Time-Critical Strike (TCS) against high priority targets, such as theater ballistic missile (TBM) launchers and mobile air defense units, accentuates all the problems of kill chain execution. Intelligent agent-based automation lets decision-makers rapidly and fully exploit distributed information, analysis, and strike assets to gain the knowledge advantage within the time cycle of a mobile launcher “shoot and scoot.” Lockheed Martin ATL (LM ATL), funded by the DARPA Control of Agent Based Systems (CoABS) program, has set out to demonstrate that a decision support capability based on interoperable intelligent agent technology accelerates and improves human decisions in TCS operations. LM ATL is participating in the USN Fleet Battle Experiment (FBE) series, inserting the Cooperating Agents for Specific Tasks (CAST) agent systems into the experimental digital fires network. In FBE-E (Spring 1999), cooperating CAST agents monitored events from multiple intelligence sources. While individual pieces of information may seem insignificant, “scout” agents collect corroborating evidence and perform threat assessment on the correlated picture. Such an exhaustive search cannot be performed without automation, and no other automation tool has the flexibility and efficiency of agents to deal with disparate information from legacy systems. In FBE-H (Summer 2000), CAST collected target-relevant information and images from various shipboard C4I systems, highlighting facts of critical importance, such as no-strike targets. CAST presented operators with a wealth of decision-ready information at the click of a button. Strike operators routinely used CAST to increase their situational awareness across the warfare grids. CAST agent technology readily provides autonomy, persistence, and efficiency to monitor dynamic information channels. CAST agents extract information and images from relational databases, Web pages, and email message streams.

Upload: phamdiep

Post on 14-Feb-2017

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A1

Distribution Statement A: Cleared for Public Release; distribution is unlimited.SPAWAR Security & Policy Review control number: SR-2001-098.

CAST Agents: Network-Centric Fires UnleashedMartin O. Hofmann, Daria Chacón, Gerard Mayer, Kenneth R. Whitebread

Lockheed Martin Advanced Technology LaboratoriesCamden, NJ 08102

{mhofmann, dchacon, gmayer, kwhitebr}@atl.lmco.com

James HendlerChief Scientist, DARPA/ISO

Arlington, VA [email protected]

Abstract

Time-Critical Strike (TCS) against high priority targets, such as theater ballistic missile(TBM) launchers and mobile air defense units, accentuates all the problems of kill chainexecution. Intelligent agent-based automation lets decision-makers rapidly and fullyexploit distributed information, analysis, and strike assets to gain the knowledgeadvantage within the time cycle of a mobile launcher “shoot and scoot.”

Lockheed Martin ATL (LM ATL), funded by the DARPA Control of Agent BasedSystems (CoABS) program, has set out to demonstrate that a decision support capabilitybased on interoperable intelligent agent technology accelerates and improves humandecisions in TCS operations. LM ATL is participating in the USN Fleet BattleExperiment (FBE) series, inserting the Cooperating Agents for Specific Tasks (CAST)agent systems into the experimental digital fires network.

In FBE-E (Spring 1999), cooperating CAST agents monitored events from multipleintelligence sources. While individual pieces of information may seem insignificant,“scout” agents collect corroborating evidence and perform threat assessment on thecorrelated picture. Such an exhaustive search cannot be performed without automation,and no other automation tool has the flexibility and efficiency of agents to deal withdisparate information from legacy systems.

In FBE-H (Summer 2000), CAST collected target-relevant information and images fromvarious shipboard C4I systems, highlighting facts of critical importance, such as no-striketargets. CAST presented operators with a wealth of decision-ready information at theclick of a button. Strike operators routinely used CAST to increase their situationalawareness across the warfare grids. CAST agent technology readily provides autonomy,persistence, and efficiency to monitor dynamic information channels. CAST agentsextract information and images from relational databases, Web pages, and email messagestreams.

Page 2: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A2

CAST agents execute according to a modular, task-oriented workflow model, whichgives them persistence and autonomy. Tasks and workflow are configurable, reusablecomponents; this minimizes time and effort to tailor CAST to new decision supportapplications. CAST technology addresses security and resource management and robustlydeals with sometimes unreliable and congested military networks.

CAST leverages the DARPA CoABS Grid, an innovative architecture and infrastructurefor on-demand system integration. The CoABS Grid gives CAST agents access tofluctuating resources on networks that change over the course of the deployment.

After FBE-H, COMSECONDFLT described CAST as “showing promise, replacingredundant manual manipulations”. At a recent, DARPA-sponsored meeting of the agentresearch community, RADM Robert Sprigg, NWDC, cited CAST in his keynote addressfor its successful application of agent technology in the FBE’s. The InteroperableIntelligent Agent Toolkit (I2AT), which we are developing under DARPA funding, willovercome a major transition hurdle; it will enable domain expert users to configure andpersonalize intelligent, agent-based automation processes.

1. Introduction

The increasingly sophisticated asymmetric tactics of overmatched opponents haveprompted research and experimentation into processes that accelerate and improve theU.S. military response. Locating and correctly identifying mobile targets is an essentialcapability in this response. The USN Navy Warfare Development Command1 (NWDC)in Newport, RI, addresses warfare innovation in terms of developing new doctrine andconcepts, by war gaming and experimentation. The Maritime Battle Center department ofNWDC coordinates the execution of Fleet Battle Experiments with the numbered fleetswhere operators and NWDC personnel jointly exercise these innovative warfareconcepts. FBE concepts and initiatives have included Network-Centric Warfare, TheaterBallistic Missile Defense, and Time-Critical Strike.

The Time-Critical Strike (TCS) concept development aims to shorten the time to detectthe fleeting threat, to decide if and how to engage it, to engage with the chosen weapon,and to assess the damage inflicted. Decision makers and supporting operators arechallenged to collect and interpret the available data and imagery, to analyze and choosecourses of action, to coordinate among the multiple operators on distributed platforms,i.e. ships, naval aircraft, and Marine land forces, and to monitor for unexpected changesin the situation. FBE architects deploy advanced systems for target detection,mensuration, weapon-target pairing, and fires coordination. However, each of thesesystems requires human operators and the TCS process results from human operatorscoordinating among each other with simple tools, such as Internet Relay Chat (IRC). Theopportunity exists to accelerate the TCS process through greater levels of automation.

FBE initiative leads team with industry and design novel system architectures thatspecifically support the topics of experimentation. This architecture is composed ofsystems ranging from experimental systems to systems near transition, such as GISRS

1 http://www.nwdc.navy.mil

Page 3: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A3

(Global Intelligence, Surveillance, and Reconnaissance System)2 to systems of record inthe Global Command and Control System – Maritime (GCCS-M), such as theModernized Integrated Database (MIDB). Interconnecting these diverse constituentsystems for the experiment requires significant ingenuity and resources. There is a needfor a more flexible approach to system connectivity and interoperability.

With the hypothesis that intelligent software agent technology will to lead to arevolutionary new model of computing beyond client-server architecture, the DARPAControl of Agent Based Systems (CoABS) program3 is investigating control andcoordination of distributed, heterogeneous agents and non-agent services. Its goal is todevelop the technologies that make possible systems of cooperating agents and agentensembles that dramatically reduce the information systems workload for the entirespectrum of military forces.

Lockheed Martin Advanced Technology Laboratories (ATL) is participating in theCoABS program under the MACOE (Multi-Agent Common Operating Environment)contract. One of our main objectives is to initiate transition of CoABS results to themilitary. Starting with a demonstration system in FBE-D, we have fielded a series ofincreasingly capable multi-agent systems in coordination with NWDC. This June, CASTsailed in FBE-I, fully integrated into the FBE system architecture. We are leveraging ourJava-based mobile agent platform EMAA4 (Extendible Mobile Agent Architecture) andour CAST (Cooperating Agents for Specific Task) application agent framework todevelop multi-agent decision support systems to support Time-Critical Strike in theFBEs.

The agent-oriented system model significantly improves the way decision supportsystems are created and fielded and how users interact with the resulting autonomous,process-aware systems. CAST applies the agent-oriented model of autonomous,distributed computing to time-critical strike decision support. CAST agents become thesecond pair of eyes and the second pair of hands of the decision maker. Like an extensionof their human user, CAST agents watch for target nomination messages, collectinformation from C4I systems and participate in collaboration channels. Unlike theirhuman user, CAST agents monitor multiple, simultaneous TCS processes with constantvigilance, dig information from cumbersome legacy system interfaces, remember tocheck for the most recent data and image updates, and keep an accurate history of allactivity. CAST accelerates strike/no-strike decisions and increases situation awarenessfor better decisions based on the right information.

CAST applies the CoABS model of dynamic system integration to time-critical strikedecision support. CAST employs the CoABS Grid, a middleware that integratesheterogeneous agent, object, and legacy systems. The CoABS Grid, based on Sun’sJini™ network technology, supports formation of a dynamic system of systems, where

2 GISRS has since transitioned into the GCCS-M as GISR-C (GCCS-M Intelligence Surveillance andReconnaissance Capability).3 http://www.darpa.mil/ito/research/coabs/index.html4 Susan McGrath, Daria Chacon, and Ken Whitebread, Intelligent Mobile Agents in the Military Domain,Proceedings of the Autonomous Agents 2000 Workshop on Agents in Industry, Barcelona, Spain, 2000.

Page 4: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A4

components join and leave while the overall system continues to function. CAST can thusbe rapidly extended to interact with additional information sources.

The agent-oriented paradigm and our CAST agent platform, in particular, acceleratesystem development for specific applications and exercises. Already, CAST multi-agentsystems are much quicker to build and field than traditional, monolithic decision supportsystems. However, it still takes skilled developers to adapt CAST to a specific FBE. Ourgoal is to develop a toolkit that will let military decision makers compose and configureCAST for their personal needs with minimal recourse to contractor support. Already,CAST agents can be composed from reusable, Java bean-like tasks and conform to one ofa small number of behavior patterns. We believe that such a toolkit will be a sufficientenabler to transition agent-oriented systems into widespread use as decision supportsystems.

This paper is organized in the following sections. In section 2 we explain how CASTagents fit into the TCS process. Section 3 contains an account of the experience withfielding CAST in the FBE series. In Section 4 presents some key technical features ofATL’s agent technology and in Section 5 we propose that agent technology does notcompete with but complements other current technology trends, such as Web-enabledapplications. Section 6 summarizes the impact agent technology has on decision supportand Section 7 illustrates these points with examples of applications of agent technologyto other military processes. Section 8 concludes with a look at future developments.

2. Agents in the TCS Process

Time-Critical Strike (TCS) against high priority targets, such as theater ballistic missile(TBM) launchers and mobile air defense units, accentuates all the problems of kill chainexecution. Time-Critical Targets (TCT) must be attacked within a dwell time of wellunder an hour, nominations interrupt ongoing strike operations, multiple strikes must beplanned at the same time, and multiple sensor, analysis, and weapon assets mustcoordinate their activities. It is difficult for operators to maintain situational awarenessfor all targets and to correctly prioritize their assets.

Among the four phases of TCS, Detect, Decide, Engage, and Assess, CAST agents mosteffectively support the “Decide” phase. CAST agents cooperate with the Digital TargetFolders (DTF) provided by NWDC to help maintain situational awareness required foroptimal strike decisions over large numbers of priority targets. DTFs are a product of the“Information Knowledge Advantage” (IKA) initiative. Via a Web server, IKA publishesrelevant information ranging from commander’s guidance documents to a set of dynamicDigital Target Folders. Each target folder is designed to aggregate all the informationrequired to effectively and expeditiously strike the target. Keeping DTFs current ismanpower intensive. Before FBE-H, Second Fleet operators, for example, stated that theyhardly ever had time to check the guides on attacking hardened targets even though theyare published on a Web site. CAST agents effectively remedy this lack of informationfocused on the particular target.

The typical concept of operation for CAST agents supporting TCS, as exemplified byCAST in FBE-I, starts with the USMTF target nomination message. This messageoriginates from the GISR-C or the NFN (Navy Fires Network) operator and specifies an

Page 5: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A5

initial target location and type. One agent monitors a mailbox for such messages and,once it has processed the message, generates subordinate information agents. Informationagents access database, Web, mail, and chat servers and extract data and images in closeproximity to the nominated target. For example, one agent matches the target nominationto the no-strike target list in order to bring potential collateral damage issues to thedecision maker’s attention. CAST presents all data records and images on a Web pagefrom which the user can select the appropriate items for insertion into the DTF. A humanuser, therefore, is in charge to approve the results returned by CAST agents.

Users call up the data and images collected by CAST agents with a single button click onthe DTF Web page. All users share the same CAST results Web page and can tell whatitems have already been inserted into the DTF. Users have frequently commented onhow much easier it is for them to retrieve information with the help of CAST comparedto the often cumbersome interfaces to legacy systems, such as the MIDB. CAST has theadvantage that it is aware of the current context and the users’ information requirements.CAST can therefore anticipate the users’ needs.

Information agents assigned to dynamic data sources maintain persistent watch forchanges after the initial results have returned. For example, agents persistently monitorthe Image Product Library (IPL) for new imagery that might assist in targeting before thestrike or in bomb damage assessment after the strike.

CAST agents monitor the collaboration between operators. This is a popular feature,because the operators cannot pay attention to parallel activities of dozens of otheroperators that may later turn out to contain relevant facts. The FBE-I Fires Lead LCDRErik Burian, NWDC, has sent this email during FBE-I:

“[…] The smart agents in CAST are automatically scanning all email and chat fordialog on that particular target. […] It's an SA multiplier. […]”

A newer concept for CAST is our design to assist the decision makers in prioritizingtime-critical targets. The intense time pressure associated with short dwell time targetsleads decision makers to place the highest priority on these targets regardless ofcommander’s guidance. We have begun to add a decision support capability to CASTthat suggests target priorities based on commander’s guidance, the air tasking order, thetarget nomination messages, and enemy order of battle.

3. NWDC Fleet Battle Experiments – CAST Applications

Earlier work with the 201st Military Intelligence (MI) Brigade in Ft. Lewis, WA, hadestablished that software agents could significantly improve distributed human decisionmaking.5 In the FBE series, we were able to demonstrate the utility of agent-baseddecision support for Theater Missile Defense and Time-critical Targeting. Each of theseapplications requires collection of information from distributed, heterogeneous sources,

5 Martin O. Hofmann, Amy McGovern, Kenneth R. Whitebread, “Mobile Agents on the DigitalBattlefield,” Proceedings of the Second International Conference on Autonomous Agents (Agents ’98),Minneapolis/St. Paul, MN, May 9-13, 1998, pp. 219-225.

Page 6: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A6

focusing and interpreting the information, and integration into a human, distributeddecision process.

In FBE-E (Spring 1999), cooperating CAST agents monitored events from multiplesources.6 While individual pieces of information may seem insignificant, “scout” agentscollect corroborating evidence and perform threat assessment on the correlated picture.True to the CoABS hypothesis, we verified that an ensemble of specialist agents couldcooperate and solve this difficult problem efficiently. The problem is challenging becauseit combines aspects of data mining with data correlation and human judgment. We fieldeda set of Information Agents that mined individual data sources and posted their findingson shared Whiteboards (Bulletin Boards). There, Scout Agents attempted to satisfy thecondition sets of particular hypotheses created by Monitor Agents. Monitor Agentsmanaged competing hypotheses, minimizing redundant work. Scout Agents tasked aFuzzy Rule Inference Engine to evaluate the final set of facts and created alerts asappropriate. Figure 1 illustrates the CAST-E architecture.

During FBE-E,CAST was set up inthe Full DimensionProtection Cell.CAST operated onsimulated data thatwere synchronizedwith the main FBEevents. CASTperformed well, andat the end of theexercise, the fireslead accepted arecommendationprovided by CAST.In FBE-E, CASTdemonstrated itsability to find facts

and draw inferences using an exhaustive search that cannot be performed withoutautomation.

In FBE-F (Fall 1999), CAST demonstrated reach-back from Fifth Fleet in Bahrain tonational resources in CONUS. CAST used mobile agents to monitor data in anintelligence database simulation installed at SPAWAR, San Diego. System integrationproblems prevented CAST from playing a significant role in FBE-F. The integrationproblems pointed to the need for an open, standards-based integration approach like theCoABS Grid.

6 Daria Chacon, John McCormick, Susan McGrath, and Craig Stoneking, Rapid Application DevelopmentUsing Agent Itinerary Patterns, March 2000, Technical Report #01-01, Lockheed Martin AdvancedTechnology Laboratories, http://www.atl.lmco.com/overview/library.html.

Figure 1. CAST in FBE-E alerted to Transporter-Erector-Launcher(TEL) indications.

Page 7: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A7

In FBE-H(Summer 2000),CAST collectedtarget-relevantinformation andimages fromvarious shipboardC4I systems,highlighting factsof criticalimportance, suchas no-strike targets.Strike operatorsroutinely usedCAST to increasetheir situationalawareness acrossthe warfare grids.CAST agentsextracted facilityinformation from

MIDB and noted units and equipment located at this facility. This information influenceschoice and planning of strike missions. CAST also retrieved no-strike target informationfrom MIDB. CAST agents interacted with the Web client to ITS (Image TransformationServer) to fetch images that show the selected target location. Figure 2 illustrates theintegration of CAST into the FBE-H architecture.

We applied CoABS Grid “Wrappers” to the CAST and the DFT database. Thesewrappers publish the CAST and DTF interfaces and manage component registration inthe CoABS Grid registry. Through the CoABS Grid interconnectivity was established ondemand during system execution, between a varying number of agents and resources.

In FBE-I (18-28 June 2001), our CAST framework had matured to the point where wecould, in a matter of weeks, configure CAST to interact with all relevant data sources onthe FBE network. CAST had thus achieved its primary mission to make available, with asingle click on the CAST button, all the information necessary for rapid, effective TCS.

The next challenges are to add value to the information by assisting in its interpretationand to turn over CAST configuration to the military users. We have initiated research intoboth challenges. We fielded a prototype target prioritization capability in FBE-I, and weare developing a toolkit that supports our innovative agent system life-cycle model, seeSection 8.

The CAST target prioritization function emulates the human prioritization process. ACAST agent reads the commander’s guidance document. In FBE-I we used the Missions,Priorities, Intent of Operations (MIPO) document published by I-MEF. When a newtarget is nominated, CAST matches its location and type against the specified priorities inthe MIPO document. CAST takes into account whether the target represents animmediate threat, such as a TBM, or threatens a planned mission. The agent makes use of

Figure 2. In FBE-H CAST filled Digital Target Folders with imagery fromITS and nearby no-strike targets from MIDB.

Page 8: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A8

the AeroText information extraction tool to identify statements about relative targetpriorities. AeroText is a commercial product of Lockheed Martin Management & DataSystems. The target prioritization agent reads the upcoming strike missions from an XMLversion of the ATO both to determine their priority relative to the emergent targets and todetermine if an emergent threat, such as a mobile air defense platform, represents a threatto one of the missions. CAST displays its target priority recommendations together withits decision rationale.

In FBE-I we verified that the CoABS Grid effectively supports dynamic systemintegration. CAST relied on the CoABS Grid to locate and access all the informationsources listed above. We have developed a small set of generic Grid wrappers forrelational databases and Web interfaces. We had to construct special purpose wrappersfor complex sources, such as MIDB, that require data validation and mediation. The one-time effort to create these Grid wrappers ranged from 1 day (COP tracks) to 3 weeks(MIDB) and has quickly been amortized through reuse.

An example of the power of the CoABS Grid model offered itself when we were asked tointegrate CAST with the XML Data Mover, another tool in the IKA initiative. Weadapted one of our Grid wrappers in less than a day and the XML Data Mover developersconnected it to their application the next day. We were able to verify correct operation thesame evening across the Internet. Through this interface, CAST agents, or indeed anyother component connected to the CoABS Grid, can prompt the XML Data Mover toupdate specific records in the DFT from the MIDB. This example shows how simpleconventions supported by a powerful infrastructure simplify open systeminterconnectivity.

During the wargame that preceded FBE-I, we demonstrated how Grid-aware agentsdetect and exploit additional resources on the fly, without human intervention. CASTagents on the USS Coronado, the Third Fleet Command Ship, detected and queried asecondary, shore-based DTF as soon as it registered with the CoABS Grid. Users of Grid-enabled agent systems, such as CAST, also benefit from the inherent robustness of thismiddleware architecture. When network connectivity to the shore-based DTF was(artificially) interrupted in the demonstration, agents recovered robustly, spared theunavailable source, and continued to query the on-ship DTF.

Agent mobility proved during FBE-I to be another effective capability of CAST agents.Our CAST system operators found that the mobile CAST agents that retrieved COPtracks from the ashore server never failed, despite frequent network connectivity lossesbetween the USS Coronado and the ashore systems. CAST mobile agents utilize aproprietary mobility service that is tolerant of low bandwidth, unreliable network links.

Participation in the FBE series has proven that autonomous, intelligent agents cansubstantially improve time-sensitive decision processes.

4. LM ATL Agents: Quick in Action, Quick to Launch

LM ATL has developed a robust, configurable mobile agent framework over the last fiveyears. EMAA, the Extendible Mobile Agent Architecture, is in fact a complete mobileagent platform with support for agent launch, migration, and management at a patentedagent “dock”, and distributed event messaging that is safe despite agent mobility. The

Page 9: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A9

EMAA architecture provides the foundation for reusability and rapid, disciplineddevelopment of distributed agent applications. Predating commercial Java networkcomputing facilities, such as Jini, EMAA provides distributed resource registration andlook-up, a clean separation of mobile agent tasks and stationary resources connected tothe distributed agent docks. EMAA is a pure Java application and has been run on SunSolaris, Microsoft Windows 98 and NT, and Linux. EMAA is the core of over a dozen ofour recent agent applications.

Additional code reusability derives from the internal architecture of an EMAA agent. Tocreate an agent, the developer builds a workflow graph from a mix of reusable andspecial-purpose tasks and adds transition logic. Reusable tasks, such as the relationaldatabase query task, are configurable for the specific query, database type, and databaseinstance. Configurable tasks are Java classes that can generate bean information objectsto assist in workflow composition. Data flow between tasks through an internalWhiteboard that serves as the working memory of the intelligent agent. Persistent andautonomous behavior is readily created in this model.

Looking for further ways to accelerate and simplify construction of distributed multi-agent systems, we have identified larger patterns in our applications. CAST, for example,contains a prototypical pattern of information management agents: select an informationsource, retrieve information (periodically or event triggered), correlate and filter theinformation, and present it to a user of consumer process. Our goal is to develop toolswith which military personnel adapt such patterns to their specific requirements andnetwork environment with minimal help from contractors.

Agent applicationsdeveloped usingEMAA do notunduly burdencomputational andnetwork resources.Measurementstaken during FBE-Ishowed that thetypical size of ourCAST agents fellbetween 5 and 6Kbytes.Comparison withother agentsystems showedthat CAST agentsadd the lowestoverhead to theactual task code

Figure 3. CAST agents are small and fast. They add minimal overheadto tasks and data.

Page 10: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A10

and data7, see Figure 3. Figure 3 graphs the time it takes to move task code and databetween two host systems over a 10 MB/s link. Each agent system adds some overheaddue to agent management, but EMAA comes closest to the raw TCP transfer times.

5. Autonomous, Mobile Agents and the Web

FBE system architects frequently consider Web technology as an alternative toautonomous, mobile agent technology. In fact, agent and Web technology complementeach other rather than compete. Web technology leverages large, centralized servers thatstore all relevant information, such as the Digital Target Folders. Commercial Webtechnology, in most cases, has been developed with the assumption of reliable, highbandwidth network connections. Today and for the foreseeable future, Navy networks donot match these expectations. For example, operators on the USS Stennis avoided usingthe Digital Target Folders during FBE-I because of the time it took to load and displaythe folders across the wireless satellite link. Web technology performs most of theprocessing at the servers, which has led to server overload during heavy usage. Appletsprovide some local processing but are severely limited by the security sandbox model.

Autonomous agents distribute the computational burden among processing nodes. Mobileagents move the information processing closer to the information source, e.g. to a nodeon the same local area network.

CAST agents use Web technology in several ways. CAST docks attach local Web serverswith servlet technology to manage interactions with users and display of results on thelocal area network. CAST agents insert information in the Digital Target Folders on therequest of a user, where they become part of the central information repositories. CASTagents retrieve information and imagery from Web interfaces where they log in, followhyperlinks, and fill in forms like the human user to select the relevant data.

NWDC and LM ATL have cooperated in the FBEs to develop this complementary Web-and agent-computing model. The combination of centralized (Web) and local (agent)processing as well as shared (Web) and user-specific (agent) displays has proven to workwell in the FBEs. Under internal research and development funding we are preparingCAST agents to leverage the ontologies and XML markup of the next-generation,semantic Web.

LM ATL’s mobile agent technology follows the applicable DoD security guidance.According to the guidance document on the use of mobile code8 our mobile agents posethe moderate, manageable risk of Category 2 mobile Java code. EMAA containsmechanisms to sign code and establish trusted code sources that mitigate the addedsecurity risks of mobile agents.

7 to appear in David Kotz, George Cybenko, Robert S. Gray, Guofei Jiang, Ronald A. Peterson, Martin O.Hofmann, Daria A. Chacón, Kenneth R. Whitebread, and James Hendler. Performance Analysis ofMobile Agents for Filtering Data Streams on Wireless Networks. Mobile Networks and Applications,7(2), March, 2002.8 Assistant Secretary of Defense, Policy Guidance for Use of Mobile Code Technologies in Department ofDefense (DoD) Information Systems, November 7, 2000.

Page 11: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A11

6. Agent Technology Impact

Autonomous agent technology is the future force multiplier for command and control.With proven information management and emergent decision support capabilities, agentswill let decision makers gain situational awareness faster, choose the right course ofaction, and cooperate more effectively.

7. Other Agent Applications

Outside the USN Fleet Battle Experiments, LM ATL has successfully applied its agenttechnology to a number of military problems. Below, we list a few examples to illustratethe breadth of applicability of our agent technology.

Domain-Adaptive Information System (DARPA) provided real-time access tobattlefield human intelligence, quickly identified critical or high-payoff targets, andincreased by two orders of magnitude intelligence analysts’ ability to monitor andrecognize critical events. Agents moved across sites connected by narrow-bandwidth,radio-based packet networks separated by tens of kilometers. Agents were taskable andreactive, adapting their behavior to the progress and events triggered by the attemptedexecution of their assigned tasks

Logistics Command and Control Advanced Technology Demonstration (LogC2ATD), Army Communications-Electronics Command. Developed intelligent sentinelagents to monitor logistics plans. The LogC2 agents’ decision processes are formalized insentinel patterns that a user configures for a specific mission.

Listen, Compute, Show – Marine (DARPA Communicator) is prototyping a newparadigm in human-computer interaction, where the computer listens for informationrequests, computes user-centered solutions, shows tailored visualizations to individualwarfighters, and responds with synthesized speech. Speech understanding combined withintelligent agent technologies produces computers that respond to spoken requests forC4I information. Intelligent agents interact with the warfighter to identify explicit andimplicit tasks as well as interact with remote information servers over tacticalcommunication links to retrieve required information or to request needed equipment andsupplies.

Small Unit Operations (DARPA) — Developed a collaborative, intelligent-agentsystem for warfighter Situation Assessment and communication of threat warnings. Thesystem combined a mobile-agent infrastructure with an innovative concept for agentcollaboration—a combination that allowed agents to opportunistically exchange threatwarnings, similar to the way humans do—to provide the right user with the rightinformation. Agents leveraged a threat ontology and a fuzzy logic knowledge base todetermine threat to individual warfighters from sensor reports on enemy movement,equipment, and activity.

Hunter Standoff Killer Team (HSKT) ACTD (Army Applied Aviation TechnologyDirectorate) — Developing intelligent agents that support situation and threatassessment for the airborne battle management system. Agents monitor the completenessof the tactical picture, tap non-traditional sources for fusion into the common picture,

Page 12: CAST Agents: Network-Centric Fires Unleashed

Distribution Statement A12

manage dissemination of data between airborne decision aiding systems, and monitorexecution of battalion-level combined arms plans.

8. Future Directions

For JFCOM Millennium Challenge 2002, LM ATL is working to broaden the CASTapplication to multi-service use, including Air Force and Army C4ISR systems. We aresupporting the NWDC Expeditionary Sensor Grid initiative, providing our mobile agenttechnology and our legacy system exploitation technology. For FBE-J, we plan tointroduce additional decision support capabilities into CAST by including agentcomponents from the CoABS program, such as schedulers and planners.

Our technology development is focused on improving the agents’ ability to interpret theinformation they encounter as well as accelerating agent system assembly andconfiguration. Our hypothesis is that agents not only revolutionize decision supportcapabilities, but also the development and insertion of decision support tools. Wherecontractors used to work for years to develop stove-piped systems, we are developing theInteroperable, Intelligent Agent Toolkit (I2AT) whose agent, task, and life-cycle patternswill let military users tailor agent applications for a specific deployment.

LM ATL is on a path to introduce nimble, intelligent software agent technology intonaval C4I systems. CAST brings automation to execute portions of the decision makers’tasks faster and more thoroughly than humanly possible. Our intelligent software agentsare “integration agents” of capabilities of existing systems, exploiting the CoABS Grid.CAST applies agents where agents are better than humans, to scan large volumes of data,correlate, remember, pay attention to multiple threads, and don’t tire.

9. Acknowledgements

CAST applications described in this paper were funded under DARPA contract F30602-98-C-0162 (MACOE). ONR contract CLIN 9, TO 1, partially supported participation inFBE-I. Special thanks are due RADM Sprigg, Commander NWDC, and to PaulSchmidle, the NWDC IKA initiative lead, for supporting insertion of CAST into the IKAsuite. We owe gratitude to the NWDC FBE directors and fires initiative leads for theirsupport and insights.