SUBMARINE TECHNOLOGY SYMPOSIUM



Preparing to do Tomorrow’s Job Today - Automated Tactical and Mission Planning Assistance for the Information Age Submarine

Thomas C. Smith (1)

Peter W. Jacobus (2)

David P. Watson (1)

(1) The Johns Hopkins University Applied Physics Laboratory

(2) SONALYSTS, Inc.

Abstract

The information overload that we are experiencing in our everyday lives is being felt to an even greater degree at periscope depth. An explosion in the amount of available data together with operations in increasingly congested littoral areas threaten to overwhelm the operator who attempts to account for it all. Most will agree that there is little reason to expect a change in this trend. Ever higher needs for cooperation with both manned and unmanned platforms as well as the continuing requirement to operate in areas of high contact density will drive us toward higher bandwidth communications, both acoustic and radio frequency. In the future, there will simply be more and more information pouring in through the antennas, periscopes and hydrophones. The challenge is not to stop or resist this flow, but to avoid being buried by it.

This trend is not new, nor is its solution unknown. The quantity of raw information available to the submarine decision maker has been growing throughout the nearly 100 years of submarine operation, primarily as a result of improved onboard sensors. Corresponding increases in and degrees of processing have been necessary to keep up with and take advantage of this growth. What appears to be different today is the speed at which things are changing and the need for step, rather than incremental, improvements in processing. If we envision the submarine decision maker standing at the top of an informational pyramid, supported by ever higher levels of processing, then it is time to consider the addition of a new level --- automated tactical planning.

Automated tactical planning tools represent a “next step” in the hierarchy of information processing that offers hope to the submarine Commanding Officer and Officer of the Deck that there will be relief from information overload. These tools are no substitute for good judgment and experience, and they may never be able to make the trade-off between ship safety and mission performance. The goal is to keep the operator’s mind free to evaluate his options and avoid the need for him to sift through mountains of information. Embedded in the right applications, automated planning aids promise to provide recommendations for both immediate and longer term actions that account for all available information with speed and accuracy.

The Office of Naval Research (ONR) is sponsoring the development of a Tactical Planning Associate (TPA) for use aboard the next generation combatants. TPA is an open software architecture that integrates automated planning, monitoring, and execution algorithms in a layered system that can be directly integrated into the submarine combat system.

The focus of this paper is to demonstrate the potential for automated planning aids such as TPA to process tactical information to a degree that relieves the operator of the burden of information overload, while preserving within him the understanding and confidence necessary to make tactical decisions. We will begin with an operational concept illustrating how automated tactical planning aids can interact with related components of the combat and weapons systems, and serve the decision maker on board a 688 class submarine operating in the 21st century. A second section will highlight technical details of the TPA architecture. Finally, we will offer an approach to transition technology to an operational submarine combat system taking advantage of current trends in open Defense Information Infrastructure (DII).

1. Vignette

The Year is 2005. USS Los Angeles, recently outfitted with the latest commercial, off-the-shelf (COTS) fire control and tactical aid programs, has arrived off the coast of the Orange nation.

Peace negotiations have recently broken off with the Orange nation. Los Angeles has received Fleet Tasking to conduct area characterization of Orange’s coastline and obtain all possible information about Orange’s medium sized Naval fleet and small diesel boat fleet. Cloud cover this time of year makes aerial surveillance virtually impossible --- the SSN is our nation’s best choice.

The CO is pleased at the initial performance of the Tactical Planning Associate (TPA). In this phase of surveillance, the TPA Generative, or long-term planning assistant is assimilating sensor data from sonar, fire control, Electronic Surveillance Measures (ESM), visual, and national sensors as well as Link sources. Already, after only several days, the TPA has assisted the CO in sorting out shipping patterns, land-based radar emissions, flight schedules, and what appears to be a preferred route for Orange’s warships entering and exiting the main port. After entering the main objectives of his mission (data collection, surveillance, reconnaissance), the TPA provides recommendations to best position the submarine to maximize collection opportunities while minimizing historical exposure threats. Of course, he would have come up with the same answers himself, but he appreciates the time saved by the clever data fusion plot that the TPA presents.

Talks with the Orange nation continue to degrade and the National Command Authority (NCA) desires to obtain further information about coastal activity, although high level coastal cloud cover remains. The decision is made to air-drop a UAV (Unmanned Airborne Vehicle) from an aircraft off the coast. This will allow the SSN to pilot the vehicle and collect the required visual information. The CO again consults the TPA and is able to recommend a suitable box of water that has had no shipping contacts, air coverage or shore-based radar threats at this time of day in the last week, which should allow for continuous operation at periscope depth during the flight. The UAV is dropped from a safe stand-off distance and the SSN is able to successfully control it and gain great quantities of useful information.

As strike missions are being planned (to be executed if talks fail) the Weapons Officer is delighted that the new electronic data hub allows for all critical systems to talk together and share data. Mission planning, which used to consume many hours of manual plotting, is now done electronically by his operators. The SSN’s location, contact situation and requirements for launch are all taken into account by the Tomahawk Planning Associate. The resultant launch box is fed into the TPA Procedural, or near-term, planner, which confirms that all known constraints (course, speed, contacts and geographical) will still permit the ship to be in the box at the prescribed time, if needed. Fleet tasking changes, when they happen, are entered into the planning system and new launch boxes are drawn in short order. The inevitable last minute changes can now be accommodated with greater accuracy and timeliness.

The TPA provides recommendations for each trip to periscope depth based on the proposed objectives of the trip. Not only is the TPA able to observe the current contact situation, but it draws from historical data of shipping lanes, radar and ESM contacts, as well as predicted satellite passes. The OOD is able to assess the situation and make good, consistent choices for each trip knowing that he has maximized his chance for success.

Increased contact density drives the Los Angeles to a region where they have not operated before. The SSN detects an uncharted and long-forgotten mine field --- suddenly, sonar reports high-speed screws from the North, East and South. Startled, the Officer of the Deck (OOD) picks an evasion course and begins to turn. He confirms with a quick glance at the TPA Reactive maneuvering board screen that he is within 5∞ of the optimum evasion course. Reminded by the TPA, the OOD orders masking and evasion devices launched at the optimum positions. After successful evasion, the location of the minefield is transmitted to the Force Commander as the SSN re-positions itself for the upcoming strike against the Orange nation.

Although fictional, this scenario is not far removed from what will soon be reality. Already, electronic planning aides exist aboard SSNs. Naturally, as more are introduced they will evolve into an interconnected system, which will need to be properly managed in order to efficiently assist the CO, Weapons Officer (WO) and the OOD to plan, execute, and react. The TPA can, and will, be able to meet this challenge.

2. Layered Planning Architecture

In response to emerging SSN mission requirements and constraints, the Office of Naval Research (ONR) is sponsoring the development of a TPA tool for use aboard next generation combatants. TPA is an open software architecture that integrates automated planning, monitoring, and execution algorithms in a layered system that is intended for direct integration into the submarine combat system. This layered approach will unify alerts, responses and prioritized recommendations in order to guide the crew to perform the optimum actions in any scenario encountered. This section starts with a general overview of the system and then proceeds with discussions focused on the development approach and the specific design issues of the layered architecture.

2.1 Overview

The human mind operates and plans on at least three levels: long, medium and short term. To compare these levels in a common framework, think of financial planning. The long term would then be like planning for retirement, the medium term would be saving for a home down payment, and the short term would be allotting money for the weekly groceries. Each requires different planning patterns and strategies, but all build on each other.

In a similar vein, the three tiered layered architecture shown in Figure 1 emulated this “human like” planning behavior. It is designed to effectively coordinate dynamic control/feedback event horizons, plan decomposition, and assessment processing. The integration of real-time performance and purposefully directed future planning is enhanced by the layered design combining the best portions of traditional algorithmic approaches with the newest in Artificial Intelligence (AI) planning technologies.

Long range plans are developed in the highest layer by exploring assessments of forecasted situations. Components of these plans are subsequently implemented within the lowest layer by tailoring a real-time response for the current situation. The middle layer bridges these significantly different event-horizons and insures purposefully directed actions in dynamic situations. This is achieved by maintaining the evolving context that directly supports a rational assessment of long range Task Plans, and by developing appropriate near-term Action Plan components for execution.

The design goal is to develop an adaptable unified approach that effectively merges technologies with dissimilar assessment and response needs, and significantly different processing requirements, into a system that complements and enhances the value of the individual sub-systems. The layered approach involving both planning and action is the best scheme to achieve this objective.

2.2 Approach

The general approach is based on the Cypress planning architecture [1] which employs a plan generator and executor that share a library of available actions. The generator develops plans to achieve long term, far-reaching goals while adhering to constraints given by both the operator and generated by other portions of the TPA. In contrast, the executor is always active, constantly monitoring the tactical state and the current plan, determining alternative and parallel actions, and requesting new plans when unanticipated events change the current plan.

The development approach is to enhance the Cypress components with not only a high rate reactive layer, but also add continuous planning/control feedback between the layers. The difference in computational requirements associated with higher levels and the real-time processing of the lower levels can be effectively managed asynchronously because each layer is both proactive and responsive. This concurrent processing enables higher rate components to address the immediate situation while higher level components work on more abstract replanning and plan evolution refinements.

Layering also encapsulates data and function, forcing modularity and well defined interlayer interfaces. As with the analogous Object Oriented Programming (OOP) approach, this form of structured data/function partitioning greatly simplifies maintenance and subsequent technology adjustments. Layering also incorporates data abstraction through task reduction, thus facilitating the alignment of goal concepts with the available reasoning approach to help attain a desired or reasonable processing duty cycle. In essence, the output of the layers are the goals (inputs) for the adjacent controlled layer. Similarly, a layer’s actions and internal state provide valuable execution monitoring status to the adjacent controlling layer. In this fashion, individual layers provide both control and feedback to adjacent layers as they collectively strive to realize the highest level objectives.

A hierarchical decomposition of plans into tasks, and tasks into actions is well suited to this layered approach. While generating more manageable implementation details, however, very specific constraints, requirements, and dependencies are also imposed. In particular, adding detail to a task will generally generate specific scheduling requirements for one or more of the formulated elements. One approach to this problem is to simply abandon scheduling and let events drive the actions. This Reactive Planning approach is very effective, but has an obvious limitation. It can not schedule. Therefore, this technology is clearly inappropriate for implementing strictly ordered procedural activities. It is also inappropriate, if not impossible, to develop event driven actions that insure globally optimal solutions. Nonetheless, Reactive Planners provide reasonable solutions and real-time performance.

Consequently, an implementation objective is to spread the planning activities across the layers and utilize the most appropriate technology for each layer’s tasking. The lower layers amplify task detail while also providing faster cycle times. Each layer cooperates with the adjacent lower layer by providing control (in the form of direction and context), and feedback (status monitoring) to the neighboring higher layer by recognizing and signaling specified problem states. This teamwork provides valuable concurrent processing time. Higher layers may develop or refine a more global solution by utilizing the lower layer’s ability to furnish processing time as it renders appropriate, but not necessarily optimal, event driven behavior. In essence, each lower layer supports both faster execution and additional plan detail by narrowing the focus within the current tactical situation, while each higher layer maintains a consistent global solution by insuring that the unfolding tactical development satisfies the broader mission perspective.

2.3 Design

As illustrated in Figure 2, each of the three planning layers provides Assessment/Monitoring and Response/Control capabilities. These elements handle the layer specific planning, information sharing, and control/feedback functions that are necessary to support concurrent processing using a hierarchical decomposition of control [2]. The Assessment and Response interactions encapsulate the information analysis and action generation transforms that are embedded in the layer’s specific planning technologies. In contrast, the Monitoring and Control pairing implements the essential layer specific system level signal and data sharing aspects of concurrent processing. The World Model (i.e., Scene and Resource Servers) and Synchronization Server components also directly support the cooperative asynchronous interaction that is essential to achieve an effective coordination of concurrent planning activities.

2.3.1 World Model

The World Model spans all three planning layers and furnishes a consistent representation of the current scene and resource information as it is made available by the shipboard systems (sensors, procedures, environment, etc.). In addition, the World Model provides a practical vehicle for interlayer communications and data driven signaling by supporting the global posting of information from inside each layer.

The Scene Server provides the best estimate of the contact situation. Current chart and environmental data are also maintained. There is also a tactical data base of detailed data which enumerates and characterizes the limits of platform capabilities (including ownship), weapons, countermeasures, and sensors. This data base is able to provide platform specific characterizations as well as generalized input.

The Resource Server captures the current state for the dynamic attributes of ownship which are relevant to one or more of the planning tasks. At a minimum, this would include detailed status of the motion, weapons, counter measures, and signature.

The Scene and Resource server operates in real-time, continually updating the world model as contacts, environment, and ownship changes occur. These processes will use the open combat system services interface being developed for the New Attack Submarine (NSSN) and commercial 688 backfit combat systems (CCS Mk2/1C), greatly simplifying system integration.

2.3.2 Synchronization Server

The Synchronization Server specifically supports signal generated communications between planning layers and enables responsive interlayer interrupts. This basic signaling capability is used to develop interlayer alerts and mutual exclusion locks on the World Model data and thus provides the essential substrate needed to implement a distributed synchronization of both control flow and data access.

2.3.3 Generative Layer

The Generative planning layer furnishes the traditional planning capabilities that are responsible for developing a globally optimal solution for execution such as reconnaissance and surveillance planning. This component couples both search and non-search based reasoning with stochastic and learning approaches to provide a high level planning capability which explores alternatives and considers abstract interactions. Generally, this processing requires several minutes to hours of processing.

The Man-Machine Interface (MMI) for this planning layer provides the operator with a set of tools that facilitate the mission level planning tasks that are managed by the CO and Tactical Action Officer (TAO).

2.3.4 Procedural Layer

The essential task of the Procedural planning layer is to insure the proper unfolding of the higher level Generative plan. This is achieved by monitoring and controlling the execution progress of the lower level Reactive layer by the Generative layer’s higher level goals, constraints, operator input and other assessment criteria. This involves reacting when the observed and predicted assessments differ significantly: by providing appropriate portions or updates of the current plan to the Reactive layer for implementation, by altering the Reactive layer’s execution context, and by requesting a plan revision from the Generative planning layer. In this manner, the thread of actions through the specified global plan is insured. The construction, selection, and activation of partial plans provides this layer with the capability of executing procedures, scripts, and plans in a dynamic environment. It is therefore essential that this layer manage the pursuit of evolving goal-directed tasks while simultaneously responding to changing events within a limited time for multiple activities. The generic processing requirements of this layer are in the order of seconds to minutes.

The MMI allows the operator to monitor, manage, and selectively control the development and execution of multiple concurrent tasks. This presentation is tailored to facilitate the tactical planning tasks that are performed by the TAO and OOD.

2.3.5 Reactive Layer

The Reactive planning layer handles short-term tasks with strict deadlines in order to detect and process the most dynamic information on the World Model. This layer is responsive to unanticipated events such as an incoming hostile torpedo and provides predictable, timely actions in both normal and crisis situations. This can be achieved with an action management capability that understands and correctly handles action interdependencies and multiple concurrent goal directed resource contention conflicts. In general, this processing requires a duty cycle in the order of a few seconds.

The MMI for this layer provides the operator with a rapidly understandable explanation of resource commitments in terms of contention, limitations, and trade-offs, since this information is critical for selectively controlling actions. This component is designed to assist the Fire Control Coordinator (FCC) and WO with their planning tasks.

2.3.6 Layer Coordination

Each planning technology provides and maintains its own copy of relevant World Model information. This buffering satisfies two important needs. First, it guarantees a stable slice of data for the duration of processing. Second, it provides a reference for analyzing the progression of World Model information that is eclipsed during its processing.

The Monitoring function in each layer supplies architecture specific real-time assessment capabilities that are needed to implement, manage, and coordinate the set of concurrently processing planners. This includes matching criteria for monitoring interlayer, intralayer, and World Model status that can be used to interrupt local plan processing. The set of monitored events can be tailored and activated either internally or externally and thus provide a structured means for effectively managing a distributed set of planning assets (multiple machines or processors).

Each layer’s Control module provides the architecture specific real-time functionality that is needed to support a peer model of distributed concurrent control. This capability has been included to provide flexibility, since the current layered implementation only requires support for a less general hierarchical model as shown in Figure 2.

With this general coordination scheme, intralayer planning activities can be initiated and directly controlled by the Synchronization Server as well as the internal Monitor and Control processes, thus enabling timely responses to both the local (intralayer) and World Model (interlayer) processes. This empowers the computationally intense Generative layer to function without impacting the performance or attention of subordinate tasks in lower layers. In essence, the Generative layer is available to forecast, model, and reason about likely outcomes to achieve mission goals as it searches to develop and refine a globally optimal solution. In addition, the developed set of global plans is contrived and maintained at a higher level of abstraction and therefore less detailed, faster to generate, and less sensitive to change. Furthermore, relinquishing the implementation details to the more responsive layers facilitates the prompt handling of anticipated events with actions that are designed to precipitate a desirable response while unanticipated events are handled symptomatically and furnish valuable time for Generative refinement.

2.4 Prototype

The Tactical Planning Associate (TPA) is funded by the ONR Science & Technology Program for Undersea Weaponry under the Tactical Engagement Assessment, Planning, and Control Technology task of the Combat Control Technology Project. The goal of this effort is to prototype an automated decision support system for Command and Control (C2) operators that provides decision makers of both surface ships and submarines the ability to generate, monitor, and execute an integrated ship plan in support of offensive and defensive Anti-Submarine Warfare (ASW) operations. The operational system must furnish real-time recommendations for approach and attack (undersea weapons targeting) as well as torpedo self-defense engagements in littoral, complex, dynamic, countermeasure-rich environments where multiple hostile, friendly, and neutral platforms are present.

The overall system approach utilizes a suite of appropriate candidate technologies to prototype the previously described layered planning architecture Implementation is proceeding from the bottom up with extensible general purpose technologies, while the top down development focuses on providing useful sets of operator tools. Java and C++ are the system development languages.

A system level control panel, as shown in Figure 3, has been developed. This interface provides a convenient way to select configuration parameters and perform system management tasks. This MMI includes adjustments for both the common and unique configuration features of each of the technologies that have been included in the TPA prototype.

2.4.1 World Model

A freely distributed Inter-Language Unification (ILU) system [10] that supports the Common Object Request Broker Architecture (CORBA) has been used to formalize the Application Programming Interfaces (API) for the World Model . Interface Description Language (IDL) units and context brokers for both the Scene and Resource Server components, as indicated (shaded) in Figure 3, have been implemented. These elements provide the fundamental connectivity needed to support the TPA distributed planning approach.

2.4.2 Generative Layer

The initial Generative focus has provided a robust Path Planning capability which employs a very efficient Recursive Best-First Search (RBFS) algorithm [3] to produce an ordered set of waypoints that inscribe an optimal path through a field of predictably moving and stationary obstacles having arbitrary avoidance areas. We have also implemented a classical Universal Quantification Partial-Ordered Planning (UCPOP) algorithm [4] that uses the powerful Action Description Language [5] to define and control plan development. The UCPOP technology provides the General Planner framework and initial functionality for managing the Path Planner.

As depicted in Figure 4, the General Planner is above the other Generative functions. This is because it orchestrates the invocation of other planning elements by tasking these specialist to solve specific problems and then integrates the recommendations into a coherent solution. In the current implementation the Path Planner is the only specialist, but this toolbox of planning capabilities will expand to include other forms goal directed inferencing like scheduling, planning with uncertainty, and learning. The MMI for the Path Planner is shown in Figure 5.

The development and maintenance of an extensive selection of mission templates is a natural extension to Generative layer which can be supported by including a cased-based reasoning capability. This technology provides selective storage an retrieval of mission specific profile information that can be used to facilitate the development of new mission plans through the management and adaptive tailoring of existing plans. These profiles would include Anti-Submarine Warfare, Anti-Surface Warfare, Strike Warfare involving covertly launching Tomahawk missiles at shore targets, Mine Warfare involving both laying and navigating mine fields, Naval Special Warfare including Special Operations Forces (SOF) insertion and retrial, and Surveillance missions.

2.4.3 Procedural Layer

A leading edge Java implementation of the Procedural Reasoning System (PRS) [6] technology called Jam! [9] is used in this layer. Procedural reasoning differs from other structural knowledge representation schemes because the ordering and conditional information embedded in procedures is also preserved. This feature results in focused, goal-directed reasoning where plans are expanded as procedural routines. As new facts or goals are discovered, intentions are reassessed, and actions may be initiated, interrupted, or stopped. This reactive capability enables the system to change its focus completely and pursue new goals in response to evolving situations while maintaining continuity with the global purpose.

Procedural reasoning effectively supports the representation and dynamic application of both procedural and structural knowledge. This capability is ideal suited for capturing the operational doctrine that is contained in Naval Warfare Publications (NWPs) and specializing its all-purpose tactical recommendation for the evolving situation. Environmental considerations, weapons and countermeasure inventories, equipment status, land masses, and other relevant considerations which are not detailed in the NWP can be woven into specific PRS knowledge bases. Collections of these plans can be cataloged into mission specific libraries which provide the building blocks for implementing a cased-based Generative layer learning capability as described above.

A Jam! process, i.e. agent, contains five major functional elements. There is a local database that maintains the current set of beliefs. A library of available plans or Knowledge Areas (KA) that can be implemented to achieve a goal. An intention structure that captures the state, status, and the progress for the current set of active KAs. An interpreter that provides the control cycle needed to drive the planning process. And lastly, hooks for augmenting the basic interpreter functionality.

Multiple Jam! agents can run in parallel. An asynchronous message-passing facility is being developed that will enable communication among agents to support distributed, cooperative problem solving.

The current TPA Procedural layer contains a single Navigation agent. This agent has KA’s for managing the ordered set of waypoints that are provided by the Generative layer, as well as KA’s for effectively detecting and mitigating situations that would otherwise be handled inappropriately in the Reactive layer. The navigation agent manages execution of the Reactive layer by either altering the World Model data that drives the Reactive assessment, or by changing the set of prioritized goals and constraints that define the Reactive plan.

2.4.4 Reactive Layer

In order to address the real-time planning requirements of this layer a behavior-based approach called Influence Networks (INet) [7] was selected. An INet is an event-driven network of myopic task agents and resource managers. The agents process goals to generate influences on resource managers. These managers in turn identify conflicting influences and generate resultant actions. The INet architecture provides an efficient, robust means to coherently coordinate individual tactical agents that are competing for limited system capabilities.

Individual task agents are planning specialist that validate sets of actions by influencing the required system capabilities controlled by resource managers. Response feasibility is validated by successfully influencing managers to select actions that are consistent with the goals of its event-specific response. The managers postpone resource commitment until all the task agents have validated their plans, thus generating a locally optimal solution for the contention over their resource. For a given goal set, repeated interactions with the dynamic environment produce a reproducible set of event-specific behavior.

TPA currently implements a motion (course and speed) resource that manages contention using geometric constraint-based reasoning [8]. This extension of the maneuvering board technique provides a very powerful decision tool and visualization capability that effectively manages and presents the existing trade-offs in the motion decision space. The MMI for this planning system is shown if Figure 6.

2.4.5 Future Directions

It is planned to extend the Generative Layer capability to encompass scheduling and other forms goal directed inferencing. In addition, learning approaches that support problem-solving by abstraction of persistent procedural and contextual knowledge will also be explored. Efforts which investigate developing a Feedback Classifier to provide global mission level assessments of the World Model information, and a Contingency Planner that yields feasible alternative plans are also needed to effectively address unit and force level planning issues.

Weapons Targeting and Self-Defense procedural agent development which includes the essential ASW knowledge capture and implementation in the Generative and Reactive layers is planned and currently supported by ONR.

3. Transition Path

Transition of TPA technology into fleet operation must be an evolutionary process. The end result may be a transformation of fundamental operational concepts which cannot occur overnight. The problem is really twofold: the TPA must be closely coupled with many, independent subsystems and subfunctions of the submarine combat system; also, the system will be closely coupled with the human decision makers responsible for devising and executing various missions. An additional factor is that automated planning technology will continue to rapidly evolve over the next 10-20 years. The transitioned process must be flexible enough to quickly capture technical innovation in this area and incorporate it without wholescale redesign or replacement of a validated and familiar baseline. For all these reasons, an iterative process of technology introduction is planned for transition of TPA into fleet operation.

Figure 7 shows a potential transition schedule for TPA technology that is consistent with the introduction and early evolution of New Attack Submarine (NSSN) and commercial 688 backfit combat systems (CCS Mk2/1C) that will have the capability of hosting such a system. In fact the use of open combat system interfaces and standardized platforms and operating systems greatly facilitates the early transition of TPA system software, as all prototype development is currently being performed in accordance with these standards. The first phase of TPA introduction will focus on the problem of motion management, as this is the most mature dimension of automated mission planning at this time. This will include all aspects of route planning, to various goals and constraint criteria as well as reactive tools to recommend maneuvers in complex, dynamic situations. The primary goal of the first phase system is to establish an early baseline capability integral to the combat system, with proven capability, such that operators begin to obtain familiarity not only with the specific features of the system, but with the general concept of automated assistance at multiple mission levels. This will be crucial to acceptance of future system capabilities, which would be introduced through subsequent prototype/transition phases. As shown on the schedule, second phase capability would focus on resource planning and management. This would introduce a new dimension of planning capabilities into TPA, based on sophisticated weapons and autonomous vehicle models. A third phase would focus on threat behavior modeling, for use in engagement planning and management as well as onboard training. At this point it is assumed that system maturity and crew familiarity would sufficiently advance to allow TPA to become a trusted high level decision support component of the combat system.

References

1. Wilkins, D.E., Myers, J.D., Lowarance, and Wesley, L.P. “Planning and Reacting in Uncertain and Dynamic Environments.” Journal of Experimental and Theoretical AI, Volume 6, 1994, pages 197-227.

1. Dean, T.L., Wellman, M.P. Planning and Control. San Mateo, CA: Morgan Kaufmann. 1991.

1. Korf, R.E., “Linear-space best-first search,” Artificial Intelligence 62, pages 41-78, 1993.

1. Penberthy, J.S., Weld, D.S., “UCPOP: A Sound Complete, Partial Order Planner for ADL,” Proc. 3rd Int. Conf. on Principles of Knowledge Representation and Reasoning, pp. 103-114, October 1992.

1. Pednault, E., “ADL: Exploring the middle ground between STRIPS and the situation calculus,” Proc. Knowledge Representation Conf., 1989.

1. Lee, J., Huber, M.J., Durfee, E.H., Kenny, P.G. “UM-PRS: An Implementation of the Procedural Reasoning System for Multirobot Application.” Conference on Intelligent Robotics in Field, Factory, Service, and Space (CIRFFSS ’94), 1994, pages 842-849.

1. Tychonievich, L., Smith, T.C., and Evans, R.B. “Influence Networks: A Reactive Planning Architecture.” The Seventh Conference on Artificial Intelligence Applications (CAIA ’91), 1991, pages 354-360.

1. Smith, T.C., Evans, R.B., and Tychonievich, L. “AUV Control Using Geometric Constraint-Based Reasoning.” IEEE Symposium on Autonomous Underwater Vehicle Technology, 1990, pages 150-155.

1. Huber, M.J., “Jam! Agents for Geniuses,” Users Manual, Intelligent Reasoning Systems, . Jan 1998.

1. Janssen, B., Spreitzer, M., Larner, Dan., Jacobi, C. “ILU 2.0alpha12 Reference Manual”, Xerox Corporation, . 1997.

Biographies

Thomas C. Smith is a member of the Senior Staff at The Johns Hopkins University Applied Physics Laboratory (JHU-APL) and currently supervising the Advanced Reasoning Section of the Submarine Technology Department’s Information Technologies Group (STI). He received his B.S. from the University of Maryland (1972) and his M.S. in Computer Science from The Johns Hopkins University (1984). He has experience and interest in automated planning systems with an emphasis on reactive planning technologies. His experience with the design and development of complex decision support system started at the National Security Agency and extends over the past fourteen years including efforts at both Lockheed Martin Corporation and JHU-APL. While developing automated planning approaches and prototypes for Autonomous Underwater Vehicles and submarine Command and Control applications he co-invented the Influence Network reactive planning architecture. His email address is tom.smith@jhuapl.edu

Peter W. Jacobus received his B.S. (Electrical Engineering) in 1985 from the University of Southern California and his M.S. (Engineering Acoustics) in 1991 from the Naval Postgraduate School. From 1985 to 1997, he served in the U.S. Navy Submarine Force aboard USS Chicago (SSN-721) and USS Santa Fe (SSN-763) as well as on a staff tour at Submarine Group Seven, Yokosuka, Japan. He currently serves as a Principal Analyst for Sonalysts, Inc. at their Columbia, MD office, working in support of the SSBN Force Security Assurance Program. He has co-authored several papers for the Journal of the Acoustical Society of America. His email address is jacobus@.

David P. Watson graduated from Hampden-Sydney College in 1980 with a B.S. in mathematics. In 1987 he was awarded a M.A. in applied mathematics from California State University, Fullerton. He is currently a project manager and assistant supervisor for the Information Technologies Group of the Johns Hopkins University Applied Physics Laboratory Submarine Technology Department. Over the past 15 years, Mr. Watson has been involved in the research and development of intelligent computing systems for sonar signal analysis, submarine signature management, and autonomous vehicle control. His primary current interest is in architectures for integrated tactical planning and real-time control. His email address is dave.watson@jhuapl.edu.

[pic]

Figure 1: Layered Planning Architecture

[pic]

Figure 2: Layered Information and Control Processing

[pic]

Figure 3: Tactical Planning Associate (TPA) Functional Architecture

[pic]

Figure 4: TPA Transition Schedule

[pic]

Figure 5: TPA System Control and Configuration MMI

[pic]

Figure 6: TPA Path Planner MMI

[pic]

Figure 7: TPA Reactive Maneuvering Board MMI

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download