ARMY AVIATION SITUATIONAL AWARENESS THROUGH INTELLIGENT AGENT-BASED
by user
Comments
Transcript
ARMY AVIATION SITUATIONAL AWARENESS THROUGH INTELLIGENT AGENT-BASED
ARMY AVIATION SITUATIONAL AWARENESS THROUGH INTELLIGENT AGENT-BASED DISCOVERY, PROPAGATION, AND FUSION OF INFORMATION Stephen Jameson and Craig Stoneking Lockheed Martin Advanced Technology Laboratories Camden, New Jersey [sjameson, cstoneki]@atl.lmco.com ABSTRACT The Army Aviation community is promoting the development of technologies and systems that support effective on-themove command of airborne and ground-based maneuver forces through shared situation awareness and decision aiding technologies. The operational concepts for these technologies and systems are characterized by the extensive use of mobile sensing systems, unmanned platforms, and decision aiding systems in the forward elements of the combat force, with the goal of providing mobile commanders with the improved situational awareness that results from a shared common operational picture of the battlefield. In support of these objectives, Lockheed Martin Advanced Technology Laboratories (ATL), under contract to the Army, is combining three ATL-developed technologies, Multi-Sensor Data Fusion, Intelligent Information Agents, and the Grapevine information sharing architecture, in an innovative way to provide the Shared Situation Awareness capability for ongoing Army programs in this area. In this paper, we describe the three technologies, their application in current and future work, and present results of simulation testing showing the benefits of Grapevine-enabled Distributed Data Fusion. INTRODUCTION The Army Aviation community is promoting the development of technologies and systems that support effective onthe-move command of airborne and ground-based maneuver forces through shared situation awareness and decision aiding technologies. The operational concepts for these technologies and systems are characterized by the extensive use of mobile sensing systems, unmanned platforms, and decision aiding systems in the forward elements of the combat force. This work is exemplified by the Airborne Manned/ Unmanned Systems Technology Demonstration (AMUSTD) and the Hunter Standoff Killer Team (HSKT) Advanced Concept Technology Demonstration (ACTD) programs, led by the U.S. Army Aviation Applied Technology Directorate (AATD). AMUST-D and HSKT are aimed at providing airborne warfighters and mobile commanders with the improved situational awareness that results from the cooperative construction of a shared common operational picture of the battlefield, through the sharing and fusion of information from all exploitable sensors and intelligence data sources that bear on the battlespace. In support of these objectives, Lockheed Martin Advanced Technology Laboratories (ATL), under contract to AATD, is combining three ATL-developed technologies, Multi-Sensor Presented at the meeting of the American Helicopter Society Forum 58, Montreal, Canada, June 11-13, 2002. Copyright © 2002 by the American Helicopter Society, Inc. All rights reserved. Data Fusion, Intelligent Information Agents, and the Grapevine information sharing architecture, in an innovative way to provide the shared situation awareness that is key to the success of the AMUST-D and HSKT programs (Figure 1). This paper describes the three technologies and their application in AMUST-D and HSKT. We also describe work that has been done to gather data on the effectiveness and benefits of these technologies in promoting situation awareness among a distributed team of platforms. We present the results of this work and describe resulting conclusions about the potential operational benefits of these technologies. Figure 1. AMUST/HSKT Shared Situation Awareness architecture. APPLIED TECHNOLOGIES The Artificial Intelligence laboratory at Lockheed Martin Advanced Technology Laboratories (ATL) has been developing Information Fusion and Situation Awareness capabilities for more than 10 years, including the Army’s successful Rotorcraft Pilot’s Associate (RPA) program. Our recent developments in Intelligent Information Agents have given us an additional powerful technology for information dissemination, retrieval, and monitoring on the digital battlefield. The Grapevine Information Dissemination architecture, a specialized application of intelligent agents, was developed to support opportunistic exchange of relevant data between distributed forces over bandwidth-limited data links. The individual technologies are described in the following section. Multi-Sensor Data Fusion From 1993 to 1999, ATL participated in the Army’s Rotorcraft Pilot's Associate (RPA) Advanced Technology Demonstration program, sponsored by AATD. ATL developed the multi-sensor Data Fusion system [1] that provides a common fused track picture to the RPA pilot and the RPA decision aiding systems. In the RPA Data Fusion system, data representing as many as 200 battlefield entities, from 14 different types of onboard and offboard sensors, is correlated and fused in real time into a consolidated picture of the battlespace. The RPA system, including ATL’s Data Fusion system, was successfully flight demonstrated on an AH-64/D in August 1999. The Data Fusion system (Figure 2) consists of four main elements. SourceSpecific Input Modules JSTARS Input Module Core Fusion Process Fusion Dispatch Fusion Control Client-Specific Output Modules Track Fusion Kernel CORBA Output Module MTI Fusion Kernel priate Kernel to process the data set. The Fusion Control module monitors the performance and resource usage of Data Fusion, and applies control measures such as prioritization or down-sampling of input data in cases where the Data Fusion process begins to exceed resource limitations (such as memory or CPU usage) or to fail to meet timing requirements. This separation of the top-level control from the fusion algorithms allows the Data Fusion system to be configured readily to meet different performance and resource requirements in different applications. A set of Input and Output modules manages the real-time interfaces with the sensor systems and the other components of the RPA system. The input modules read input from the sensor systems at a sensor specific rate, using a sensorspecific input protocol, and pre-process the input using tailored sensor-specific routines into a common intermediate input format and information content. The output modules take the fused trackfile and output it to a client, in a clientspecified format and using a client-specific protocol. The modular, object-oriented Data Fusion software architecture, including the use of a common intermediate data input format, permits a single core body of Data Fusion code to be easily adapted to multiple input and output formats and requirements, facilitating portability. On AMUST-D, two different versions of the Data Fusion system will be deployed on the Longbow Apache and A2C2S Blackhawk helicopters. Both versions will contain the same common fusion core, coupled with platform- and sensor-specific input and output modules. A Track Management module stores all track data and maintains the relationships among a set of track databases, one for each sensor providing input, and a Central Trackfile that stores the fused picture. Additional databases also provide access to a variety of data about platform, sensor, and weapon characteristics used by Data Fusion. Figure 2. ATL’s real-time multi-sensor data fusion system. A set of Fusion Kernel modules performs the heart of the correlation and fusion processing. As with the input and output modules, the modular nature of the fusion process makes possible the encapsulation of algorithms tailored to a specific input data type into a kernel module. As input data sets are received, the appropriate Kernel is applied to that data set, with the resulting output passed to the Track Management module to update the fused trackfile. A top-level control structure, including Fusion Dispatch and Fusion Control modules, controls the application of fusion algorithms to the input data and ensures that the system meets real-time and resource requirements. The Fusion Dispatch module evaluates incoming data, determines which algorithm set—embodied in a fusion Kernel module— should be applied to fuse the data, and dispatches the appro- The general functional flow followed by each kernel follows a similar set of steps. Figure 3 illustrates the steps followed for MTI (Moving Target Indicator) data. The Prediction function performs time-alignment across the sets of data being fused to ensure that valid comparisons can be made. The Clustering algorithms break down the battlespace into geographically distinct clusters to limit the computational Grapevine Input Module JCDB Input Module Group Fusion Kernel Intel Fusion Kernel Track Management Shared Memory Output Module SENSOR 2 CLASS: AIR DEFENSE Entity Air Fused Common Picture at time t Prediction: Sensor Data received at time t+1 and time-aligned Clustering: Data grouped into geographic clusters SENSOR 1 CLASS: TRACKED Land Wheeled Tracked Armor Resulting Class: TRACKED AIR DEFENSE Artillery Air Defense Tracked ADU Wheeled ADU ZSU-23 Resulting ID: ONE OF THESE 4 2S-6 Cost Functions: Similarity Metric applied to clusters Assignment: Associations formed based on cost values Fusion: Algorithms produce updated common picture at time t+1 Support SA-13 SA-15 Figure 3. Data fusion kernel functional flow. Figure 4. Example of class fusion in RPA data fusion. complexity of the following algorithms. Cost Functions operate on each cluster to compute a matrix of composite similarity values between each input sensor data item and the candidate fused tracks. The Assignment function uses the optimal JVC (Jonker-Volgenant-Castanon) algorithm to compute matches between sensor data and fused tracks. Once the matches are identified, Fusion algorithms are applied to update the state of the fused trackfile based on the associated sensor data. context, implies that the agent is imbued with some degree of knowledge of its environment or subject domain that allows it to make decisions that affect its behavior in response to its changing environment or the problem at hand. Many applications make use of mobile agents, that are able to travel between nodes of a network to make use of resources that are not locally available. One of the key algorithmic advances of the RPA Data Fusion system was its ability to effectively combine the processing of kinematic (position and velocity) information with the processing of Class (vehicle type), ID (specific vehicle information), and IFF (friend/hostile) information. This processing takes place during the comparison of sensor data with fused tracks, by the Cost Functions, and during the updating of the fused trackfile with associated sensor data by the Fusion algorithms. The Class Cost Function and Fusion algorithms compare and combine Class and ID information expressed in a class hierarchy (Figure 4). Each sensor report or fused track has a class representation that specifies the confidence of each node in the hierarchy. A set of Modified Bayesian Evidence Combination algorithms developed by ATL is used to compare, combine, and summarize this information. ATL’s work in this area [2] represented a major advance in Data Fusion technology. Intelligent Agents for Information Retrieval and Dissemination Since 1995, ATL has been developing technology for Intelligent Information Agents to support dissemination, retrieval, and monitoring of information on the digital battlefield [3]. An Intelligent Agent is a persistent software construct that is able to interact with its environment in order to perform tasks on behalf of the user. Intelligence, in this ATL has developed the Extendable Mobile Agent Architecture (EMAA) [4] that provides an infrastructure for the deployment of light-weight intelligent mobile agents. An agent is launched at a processing node with a set of instructions contained in an itinerary, a control construct that permits highly flexible control over agent behavior. Based on conditions or information it encounters, the agent may need to migrate to another node to continue performing its task or locate needed information. EMAA makes use of the portability inherent in the Java™ language to migrate the agent from its current processor to the target platform and execute it on that processor. The original applications of EMAA involved the use of mobile agents for search, retrieval, and dissemination of intelligence information in battlefield networks. Later applications exploited persistent Sentinel Agents for monitoring data in distributed information systems to alert a user or client when certain conditions or events occurred. In order to meet the challenge of supporting information pull as well as push, we began in 1999 to investigate the use of Intelligent Information Agents to provide information to supplement the data fusion process [5]. We focused on the problem of improving the Common Tactical Picture (CTP) by identifying areas in the fused track picture that could or should be improved through the application of additional data from other non-reporting sources. An example of such an application is illustrated in Figure 5. In this system, the output of Sensor Data Fusion is collected Fusion Input Interface Data Matching Areas of Interest Input From Multiple Sensors JCDB Data Fusion Information Agent Area Where No data is present Area of insufficient Accuracy or Latency Common Tactical Picture Sentinel Agent Figure 5. Sentinel Agent used to augment CTP. to form the basis of the CTP. A persistent Sentinel Agent examines the CTP to determine areas where additional information is needed. This analysis is based on several criteria, including: (1) areas in which data accuracy or latency does not meet requirements, possibly due to reliance on a source with high latency or large positional error; and (2) areas where no data are present, but where tactical requirements, expressed in the tactical plan, indicate a need for information. When the Sentinel Agent identifies a need for additional information, it dispatches an Investigation Agent to search for the needed information in a remote data source, such as the All-Source Analysis System (ASAS). The results of the investigation are converted into a format usable by Data Fusion, and passed as input to Data Fusion for incorporation into the CTP. This approach has been investigated in an internal research and development program and integrated into the ACT II Battle Commander’s Decision Aid at the US Army Air Maneuver Battle Lab (AMBL) at Ft. Rucker, Alabama. The implementation of the Grapevine architecture (Figure 6) builds upon our previous work combining multi-sensor Data Fusion with intelligent agents. Each node in the architecture contains a Data Fusion process that fuses locally obtained data (from local sensors and data sources) and data received from other peer nodes. The Grapevine Manager at each node manages the interchange of data with peer nodes. For each peer node, it contains a Grapevine proxy agent that represents the information needs and capabilities of that peer node. As the sensors or other sources on the platform generate local information, each grapevine agent evaluates that information against the needs of the peer platform it represents for factors such as: • Sensor type: Data from remote sensors, e.g. JSTARS, is sent only if the recipient does not already have access to that data. • Mission: For example, the peer platform’s mission may or may not require the propagation of friendly tracks. • Location: The peer platform may only need information within a geographic or temporal/geographic radius. • Coverage: The peer platform may need information from beyond its own sensor platform. Grapevine Node 1 Grapevine Node 2 Data Fusion System Data Fusion System Grapevine Manager Grapevine Manager Proxy Agent 2 Proxy Agent 1 Proxy Agent 3 Proxy Agent 3 Grapevine Node 3 Data Fusion System Grapevine Manager Proxy Agent 1 Legend Grapevine Data Dissemination The Grapevine architecture [6] was originally developed by ATL for use on DARPA's Small Unit Operations (SUO) program. It makes maximum use of bandwidth for information sharing by providing each node with a description of the information needs of its peers, so that each node can selectively transmit only those bits of information that it understands to be of real value to its neighbors. By sharing relevant sensor data, each participant can build a common tactical picture that is consistent between participants, and is as complete as the sum of all participants’ information sources can make it. Proxy Agent 2 Exchange of Information Needs and Capabilities Exchange of Sensor Data based on Needs Figure 6. Grapevine data dissemination architecture. In addition, the Grapevine agents are aware of the processing and bandwidth limitations of the peer nodes and communication links. Data identified as relevant to a peer node based on the above criteria may be down-sampled or prioritized to meet resource limitations. Each Grapevine agent propagates the needed information to the peer platform it represents, providing an intelligent push of data through the network. At the same time, the Grapevine Manager has a representation of the local platform’s information needs and capabilities, expressed in terms of available sensors and data sources, mission, location, and sensor coverage. A Sentinel Agent within the Grapevine Manager monitors the local fused picture to identify information needs not met by the local picture. Based on this, it sends out updated configuration data for the local platform to the Grapevine Manager on peer platforms. This is used to update the Grapevine Agents on the peer platforms that represent the local platform. This propagation of information needs effects an intelligent pull of data to meet the changing information needs of the local platform. There are several distinctive features of the Grapevine Architecture. First, it is a peer-to-peer network. Propagation of data occurs between peer nodes in the network, although in practice this would probably be implemented as an extension to some form of hierarchical command and control system. Second, propagation is needs based—peer-to-peer data propagation includes only data known to be of use to the recipient node, thus limiting the required processing and bandwidth. Third, the architecture is extensible. It can accommodate the addition of peer nodes merely by reconfiguring nearby nodes to reflect the addition of the new nodes. Fourth, it is survivable — there is no single point of failure. Since, in general, each node will have multiple peers, data can spontaneously reroute around missing nodes, and thus the loss of any single node will only result in the loss of the data sources local to that node. on the AH-64D Longbow Apache aircraft, and the Mobile Commander’s Associate (MCA) to be installed on the UH60 Blackhawk aircraft also equipped with the Army Airborne Command and Control System (A2C2S). Both systems include Data Fusion to provide a fused picture for Situation Awareness (Figure 7), and both include capability to provide Level 4 control of an Unmanned Air Vehicle (UAV), with both waypoint-level control of the UAV from the aircraft and direct feed of UAV sensor data to the aircraft. In addition, the WA provides decision aiding in support of the Apache pilot, including route planning and attack planning, while the MCA provides decision aiding in support of a maneuver commander, including Situation Awareness Display, team route planning, and plan monitoring. UH-60 Blackhawk Commander Mobile Commander’s Associate (MCA) Decision Aid APPLICATION IN AMUST-D/HSKT On the in-progress AMUST-D program and its successor the HSKT ACTD, ATL is developing the Shared Situation Awareness capability that will support the other functions of Pilot and Commander Decision Aiding and Manned/Unmanned teaming. AMUST-D is developing two Decision Aiding systems: the Warfighter’s Associate (WA) to be used Pilot A2C2S/ JCDB Warfighter’s Associate (WA) Decision Aid Intelligent Agent Data Discovery Data Fusion Grapevine Data Link Interface The result of this capability is to permit, in the face of stringent bandwidth and processing constraints, the creation of a Common Relevant Picture (CRP) across all participating platforms. The CRP is a shared picture of the battlefield, with all participants having a consistent view of the world, and each participant seeing that portion of the picture as it is relevant to their needs. In the case of infinite processing and bandwidth capabilities, this can scale to become a true Common Operational Picture, with all participants seeing the same complete picture. In the case of significant limitations on the ability to exchange and process information, as is the case now and for the near future, the intelligent dissemination capability of the Grapevine ensures that all participants receive the most relevant information. AH-64D Longbow Apache External Sources Link-16 IDM Data Fusion Apache Sources Grapevine Data Link Interface Figure 7. Shared Situation Awareness architecture in AMUST-D On the WA aircraft, Data Fusion will receive and fuse data from the Apache onboard sensor suite, from teammate aircraft, from UAV’s under control of the WA, and from offboard sources such as the Joint Surveillance and Targeting System (JSTARS). On the MCA aircraft, Data Fusion will receive and fuse data from UAV’s under control of MCA and from offboard sources such as JSTARS. In addition, the MCA will include the Intelligent Agent-based Data Discovery system. This will retrieve relevant blue force and red force entity data from the Joint Common Database (JCDB) in the A2C2S system and provide it to Data Fusion for incorporation in the fused picture. Data Discovery will also augment fused tracks generated by Data Fusion with additional information available from the JCDB, such as plan and status information in the case of friendly entities and sensor and weapon capability information in the case of hostile entities. The Grapevine agent system used in AMUST-D represents a specialized implementation of ATL’s intelligent agent technology in two ways. First, the Grapevine agents are implemented in C++ rather than in Java, to facilitate deployment in operational systems with stringent performance and resource requirements. Second, the Grapevine agents are being adapted to operate over non-TCP/IP networks, to facilitate use in existing tactical data links. On AMUST-D, the grapevine implementation uses the AFAPD message set over the Improved Data Modem (IDM) link to exchange data between peer aircraft. We are currently developing a prototype of this IDM implementation, and in the spring and summer of 2002 we will be conducting experiments to validate the capability of the Grapevine to support Distributed Data Fusion with the bandwidth limitations and message set imposed by the IDM. 50x50 km Battle Area 49 Randomly Distributed Targets Sensor Sensor Platform Platform West West Time Time == 00 PERFORMANCE STUDY In order to demonstrate the performance cost-benefit tradeoffs of employing Data Fusion and the Grapevine, we conducted a simulation-based experiment in which a scenario was played out without Data Fusion, with just Data Fusion, with Data Fusion and full exchange of sensor data between platforms, and finally with Data Fusion and the Grapevine. Performance results presented in this section include data on completeness and accuracy of the resulting tactical picture as well as computational and bandwidth cost. Methodology The scenario that was used in the experiment is depicted in Figure 8. In the scenario, two sensor platforms, (e.g. helicopters, or UAV’s), fly through a 50 by 50 kilometer square battle area, throughout which are scattered 49 targets, (tanks, and other vehicles). The targets are positioned according to a uniform random distribution, throughout the battle area, and each has a random velocity between 0 and 10 meters per second. Each sensor platform carries two sensors. Both sensors on the platform have the same area of coverage, which is a circle, centered on the sensor platform, with a radius of 20 kilometers. Each sensor is assigned values that define the error inherent in its reporting of the range and azimuth of a target. Each assigned error value is defined to be four times the standard deviation of a normal distribution, giving a confidence interval of plus or minus two standard deviations, or about 95%. One sensor on each platform has error values of 100 meters in range and 5 degrees in azimuth, while the other sensor has error values of 1000 meters in range and 0.5 degrees in azimuth. The range and azimuth error values for a particular measurement act as the orthogonal axes of an uncertainty ellipse that defines an 86% confidence region for the true position of the target. Each sensor on each platform reports on each target within its coverage area at the rate of once each second, throughout the 30-minute scenario. One Near 100% Sensor Coverage Overlap Time = 1200 Sensor Sensor Platform Platform East East Time Time == 00 20 km Sensor Coverage Radius Figure 8. The performance study scenario. sensor platform starts at the southwest corner of the battle area and maintains a 30 m/s velocity to the northeast throughout the scenario. The other sensor platform starts at the southeast corner and maintains a 30 m/s velocity to the northwest. At the start of the scenario, the two platforms have disjoint areas of sensor coverage. At about 4 minutes into the scenario, the coverage areas begin to overlap. At about 20 minutes into the scenario, the coverage areas overlap nearly 100%, after which, overlap begins to decrease again. Throughout the scenario, we collected performance measurements on CPU usage, number of reports being input to Fusion, number of tracks resulting from fusion of those reports, the size of the uncertainty region of the sensor reports and of the fused tracks, and the number of sensor reports passed between sensor platforms. With these metrics, we are able to conduct a cost-benefits comparison of the options of No Data Fusion, Data Fusion, Data Fusion with Full Exchange of sensor data between platforms, and Data Fusion with Grapevine. The experiment was conducted using a pair of Sun Ultra 10 workstations, connected by a 100-megabit ethernet, which is shared with a number of other nodes. Results Fusion Versus No Fusion. Multi-sensor Data Fusion provides significant improvements in the quality of the picture available to a user at a given platform, as measured by screen clutter and accuracy of track information. 2.5 Plots per Target 2 1.5 1 0.5 0 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (seconds) Without Data Fusion With Data Fusion Figure 9. With Data Fusion, the number of display plots per target is reduced to near one. Figure 10 addresses the issue of accuracy, by comparing the average size of the uncertainty region for all the tracks over the course of the scenario, with and without Data Fusion. The uncertainty value is arrived at by taking the product of the length and width of each elliptical uncertainty region. Without fusion, the dimensions of the uncertainty region are directly related to the range and azimuth errors (expressed in meters) associated with each sensor measurement. With Data Fusion, a new uncertainty region for a track is arrived at by statistical combination of the uncertainty regions of the contributing sensor measurements. Over time, this yields tracks with much smaller uncertainty than the underlying sensor data. The two curves, plotted on a logarithmic scale, show that, in this scenario, uncertainty regions without Data Fusion are on the order eight times the area of the uncertainty regions resulting from Data Fusion. Cost: CPU Usage. The benefits of Data Fusion come at the cost of increased CPU usage. Throughout the course of the experiment, for each Data Fusion processing cycle, we gathered the number of sensor reports being input to Data Fusion and the number of CPU seconds required to fuse them. Figure 11 shows the results as the average number of CPU seconds required by Data Fusion, plotted against the number of sensor reports input to Data Fusion. The plot shows some irregularity, due to artifacts of the sampling process, and the lack of data for some values of input report count, but nonetheless exhibits a roughly linear character. This is not unexpected. The time required to perform Data Fusion increases superlinearly with respect to the average size of target clusters. However, given constant cluster size, the performance of Data Fusion is linear with respect to the input size. Since the targets in the scenario are scattered about the battlefield according to a uniform random distribution, the cluster size remains roughly constant throughout the scenario, resulting in the linear character of the performance curve in the figure. 0.3 0.25 CPU Seconds Benefits: Less Clutter, More Accuracy. The benefits of employing Data Fusion can be expressed in terms of a variety of metrics. The two easiest to measure and present are screen clutter (plotted tracks per target), and accuracy, or the size of the uncertainty region associated with a track’s reported position. Figure 9 addresses the issue of screen clutter by comparing the number of plotted tracks throughout the scenario with and without Data Fusion. Without Data Fusion, each sensor on the platform reports separately on the target for each second of the scenario, resulting in a constant 2 plots per detected target on the display. With Fusion, the sensor reports are combined, so that with some variation, the average number of plots per detected target, over time, is just under 1.1, resulting in a clearer, less cluttered track picture. 0.2 0.15 0.1 1000000 0.05 100000 0 Uncertainty (m2) 1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191 Input Reports per Second 10000 Figure 11. Data Fusion CPU usage increases linearly with input size. 1000 100 10 1 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (Seconds) Without Data Fusion With Data Fusion Figure 10.With Data Fusion the confidence region for track position is greatly improved. Fusion With Full Data Exchange Versus Data Fusion Without Data Exchange. Significant improvement in the completeness and accuracy of the picture at any given platform can be achieved by the exchange of all available data among platforms, and the resulting fusion at each platform of all available data. 1000000 Uncertainty (m2) 100000 10000 1000 100 10 1 1 108 215 322 429 536 643 750 857 964 1071 1178 1285 1392 1499 1606 1713 Time (seconds) No Exchange Full Exchange Figure 13. Data Fusion maintains improved accuracy with full sensor data exchange. usage. Each Data Fusion process requires additional CPU time to process the sensor reports originating from the other platform. Also, each sensor report sent between platforms consumes communications bandwidth. From Figure 8, we know that, in this scenario, Data Fusion CPU requirements increase linearly with respect to an increase in input sensor reports. With this in mind, we can see the additional processing burden imposed by sensor data exchange by looking at the resulting increase in the rate of sensor reports input to Data Fusion. Figure 14 illustrates the relationship between the total number of sensor reports processed each second by Data Fusion with full data exchange, versus without data exchange. 350 300 Input Reports per Second Benefit: Better Coverage. When executing Data Fusion on a single platform, the platform’s tactical picture becomes more clear and accurate, but without inclusion of off-board information, the tactical picture is not complete. However, by sharing sensor data between sensor platforms, each sensor platform can have an effective coverage area that is the union of the coverage areas of the individual platforms thus providing a more complete tactical picture. To demonstrate, we ran the scenario again, with Data Fusion, but this time each sensor report that Data Fusion received from one of the platform’s onboard sensors was also sent across a communications link (TCP/IP sockets, in this case) to be used as input to Fusion on the other platform, as well. It is important to note that no fused tracks were sent, but only the raw sensor reports. The fusion of previously fused data presents special problems that are avoided by exchanging sensor data. Figure 12 compares the coverage of each Data Fusion platform, both with and without full exchange of sensor data. The coverage value is given by the number of targets seen by the platform, expressed as a percentage of the total number of potential targets on the battlefield (a total of 50 targets, including the other sensor platform). The figure shows that, throughout the scenario, the exchange of sensor data results in a significantly higher, and essentially identical, coverage value for each sensor platform, with the exception of a period of time around the 1200-second mark, when the coverage areas of the individual platforms overlap nearly completely. It is worth noting that the exchange of sensor data results in the two Data Fusion platforms seeing sets of tracks that are equal not only in number, but also in identity. The result is that the two platforms share a common tactical picture on which to base cooperative actions. Figure 13 shows that the benefits of data exchange do not come at the expense of accuracy, since the uncertainty in the position of the fused tracks is essentially equivalent with or without data exchange. 250 200 150 100 50 1 0.9 0 1 0.8 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 % Coverage Time (seconds) 0.7 No Exchange 0.6 0.5 Full Exchange Figure 14. Full data exchange multiplies the total computational burden. 0.4 0.3 0.2 0.1 0 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (seconds) No Exchange East Full Exchange East No Exchange West Full Exchange West Figure 12. Coverage of the battlefield is improved, for both platforms, with data exchange. Cost: Bandwidth and CPU Usage. The cost of exchanging all this sensor data comes in the form of increased resource Variability in the upper curve is attributed to communications link timing. It should be no surprise that, since every sensor report produced by a sensor is used as input to the local Data Fusion process, and is also sent across the communications link to be used as input to the other Data Fusion process, the total number of reports processed with exchange is twice the number without exchange. Figure 15 shows the communications cost of full data exchange, in terms of messages sent by each platform individually, as well as the combined total number of messages transmitted. It is important 160 140 120 100 80 60 40 20 0 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (seconds) East West Total Figure 15. Message traffic resulting from full sensor data exchange. to note that, around the 1200 second mark in the scenario, the total message traffic is near its peak, but benefit derived from the message traffic is at its minimum. At that time in the scenario, the sensor coverage areas of the two platforms are almost completely overlapped. As a result, very few of the messages convey information that the recipient does not already have from is own onboard sensors. Precious bandwidth is being wasted passing redundant information. Grapevine Versus Full Data Exchange. Nearly equivalent improvements in completeness and accuracy can be achieved, at a significant reduction in bandwidth and processing cost, through the use of Grapevine technology for intelligent dissemination of information. Benefits: Decreases Bandwidth and CPU Cost. The Grapevine mitigates the costs associated with full data exchange, while retaining the benefits, by restricting message production to just those sensor reports that provide benefit to the other platform. For instance, it was noted that in the case of full data exchange, when there is substantial overlap in sensor coverage areas between platforms, a substantial portion of the message traffic is of little or no benefit to the receiving platform, since that platform already has reports on the same target from its own sensors. Grapevine acts as a filter, to make sure that if the other platform doesn’t need a report, it isn’t sent. In this experiment, a receiving sensor platform is deemed to need a sensor report if the position of the target, as given by the report, is not currently within the sensor coverage area of the receiving platform. In addition, in this experiment, Grapevine further reduces resource burden through a down sampling strategy in which the reports from every other reporting cycle of each sensor are simply discarded by Grapevine. As a result, only half of those reports that would have been deemed to be needed by the receiving platform are actually sent. picted by Figures 16 and 17 respectively. Figure 16 shows the impact on message traffic by comparing the total number of messages exchanged with full data exchange, versus with the Grapevine. At the start of the scenario, when there is little sensor coverage overlap between platforms, the message traffic is cut in half by Grapevine due to the 50% downsampling. As the scenario continues, sensor coverage overlap grows, and the gap between full exchange and Grapevine widens due to the reduction in the need for sharing, as determined by Grapevine. Around the 1200 second mark, when sensor coverage overlap is near 100%, Grapevine determines that neither platform has sensor reports that are needed by the other, and so message traffic falls to near zero. As the scenario continues, sensor overlap begins to decrease again, Grapevine perceives an increase in benefit to be had from message exchange, and message traffic increases. Figure 17 shows the impact of Grapevine on CPU usage by comparing the total number of sensor reports processed by Data Fusion with full exchange versus Grapevine. The reduction in the number of input reports by Grapevine is directly attributable to the difference in message traffic between the two approaches. The idea, again, is that computational cost is only increased by Grapevine when there is a perceived benefit. 200 180 Total Messages per Second Messages Sent per Second 180 160 140 120 100 80 60 40 20 0 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (seconds) Full Exchange 350 300 250 200 150 100 50 0 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (seconds) Full Exchange The benefits of this approach can be seen by the impact it has on communication bandwidth and CPU usage, as de- Grapevine Figure 16. Use of the Grapevine reduces bandwidth cost over full data exchange. Input Reports per Second 200 Grapevine Figure 17. Grapevine reduces computational cost by reducing input reports from other platforms. To complete the comparison of Data Fusion with Grapevine versus Data Fusion with full data exchange, we need to also look at what effect the Grapevine has on the benefits that were provided by Data Fusion with full exchange. Figure 18 shows that the benefits of greater coverage that were provided by full exchange are nearly unaffected by use of the Grapevine in place of full exchange. Figure 19 shows that while there is some increase in error using the Grapevine, the level of error is still significantly lower than if Data Fusion were not used. Furthermore, it is important to realize that any increase in average error is due almost entirely to the Grapevine’s selective reporting on those targets that would not be seen at all by the sensor platform without some form of sensor data exchange. 1 0.9 0.8 % Coverage 0.7 To summarize: The use of Data Fusion offers the benefits of lowering display clutter and increasing positional accuracy at a modest computational cost. Exchanging sensor data between sensor platforms offers the benefits of increased sensor coverage to each platform and a shared tactical picture between platforms. Using intelligent, selective data exchange, such as that provided by Grapevine, can mitigate the resource usage costs incurred by full data exchange, with minimal cost tradeoff in the form of coverage and/or accuracy. A variety of alternative strategies may be employed by the Grapevine for intelligent selective data dissemination. Future work is planned to study additional Grapevine data dissemination strategies, as well as techniques for tailoring these strategies to provide maximum operational benefit in the face of dynamic and widely varying resource constraints, environmental challenges, and operational goals 0.6 CONCLUSIONS 0.5 0.4 0.3 0.2 0.1 0 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (seconds) Full Exchange, East Grapevine, East Full Exchange, West Grapevine, West Figure 18. Use of the Grapevine does not reduce battlefield coverage. 1000000 Uncertainty (m2) 100000 The development of Situation Awareness on the digital battlefield, critical to the support of mobile command of distributed forces, faces numerous challenges. It requires the ability to integrate information from all available sources and to share information between forces to the maximum extent possible, yielding a Common Relevant Picture. In this paper, we have described three technologies developed at Lockheed Martin Advanced Technology Laboratories that enable the creation of the CRP: • Real-Time Multi-Sensor Data Fusion • Intelligent Information Agents • Grapevine Information Dissemination 10000 1000 100 10 1 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601 1701 Time (seconds) Full Exchange Grapevine Figure 19. Uncertainty in target position remains low with Grapevine. Cost: CPU usage. The cost of employing the Grapevine is the computational cost of applying the filtering criteria to sensor reports that are candidates for sharing. While this cost was not measured in this experiment, it is expected that the computational cost of applying filtering to a sensor report at the transmitting end is small compared to the cost of fusing a sensor report at the receiving end. We have presented quantitative results validating the following claims about the value of these technologies: • Multi-sensor Data Fusion provides significant improvements in the quality of the picture available to a user at a given platform, as measured by screen clutter and accuracy of track information. • Significant improvement in the completeness and accuracy of the picture at any given platform can be achieved by the exchange of all available data among platforms, and the resulting fusion at each platform of all available data. • Nearly equivalent improvements in completeness and accuracy can be achieved, at a significant reduction in bandwidth and processing cost, through the use of Grapevine technology for intelligent dissemination of information. We have described ongoing work intended to provide further validation of the benefits obtained through the use of the Grapevine technology using actual tactical data link hard- ware, software, and message sets. Finally, we have described work currently in progress under the AMUST-D program that will lead to the deployment of these technologies in flight tested decision aiding systems, with a full scale Military Utility Assessment (MUA) to be performed under the HSKT program. Based on the presented results and the ongoing work, we are confident that these technologies will form the foundation of the shared situation awareness capability that will support mobile commanders in all elements of the digital battlefield of the future. ACKNOWLEDGMENTS This research was partially funded by the Aviation Applied Technology Directorate under Agreement No. DAAH10-012-0008. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation thereon. A discussion of the combination of these three technologies in support of the AMUST-D and HSKT programs has not been previously published. DISCLAIMERS The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Aviation Applied Technology Directorate or the U.S. Government. REFERENCES [1] Malkoff, D. and Pawlowski, A., “RPA Data Fusion,” 9th National Symposium on Sensor Fusion, Vol. 1, Infrared Information Analysis Center, pp. 23-36, September 1996 [2] Hofmann, M., “Multi-Sensor Track Classification in Rotorcraft Pilot's Associate Data Fusion” American Helicopter Society 53rd Annual Forum, Virginia Beach, Virginia, April 29 - May 1, 1997. [3] Whitebread, K. and Jameson, S., “Information Discovery in High-Volume, Frequently Changing Data,” IEEE Expert/Intelligent Systems & Their Applications, Vol. 10, (5), October 1995. [4] Lentini, R., Rao, G., and Thies, J., “EMAA: An Extendable Mobile Agent Architecture,” AAAI Workshop, Software Tools for Developing Agents, July 1998. [5] Pawlowski, A. and Stoneking, C., “Army Aviation Fusion of Sensor-Pushed and Agent-Pulled Information,” American Helicopter Society 57 th Annual Forum, Washington DC, May 9-11, 2001. [6] Jameson, S.M., “Architectures for Distributed Information Fusion To Support Situation Awareness on the Digital Battlefield,” Fourth International Conference on Data Fusion, Montreal, Canada, August 7-10, 2001