3D SLAM: Simultaneous Localization and Mapping in 3D Scott Robbins, Principal Investigator Problems:
Operations in the cluttered urban and interior settings pose unique challenges to unmanned vehicles. Insufficient global positioning, inaccurate pose estimation, and video sensors with narrow fields of view all lead to a lack of situational awareness for both machine and human. Exploiting 3D sensing with simultaneous localization and mapping can begin to address these problems.
Objectives:
We will investigate techniques for 3D sensor exploitation through real-time simultaneous localization and mapping suitable for use on small UAVs and UGVs operating in urban canyon and interior environments. By providing both human and machine operators with a synthesized third person "3D mosaic" and alternate localization to GPS, situational awareness can be improved in these critical areas.
Activities:
As part of this project's activities, we will directly collaborate with the AFRL Munitions Directorate's research into "Alternate Navigation" to develop a complete localization and mapping system for unmanned vehicles. Research work will be executed with a 6-DOF [degrees of freedom] platform based on a research UGV. Effective 3D sensor exploitation will be demonstrated under simulated urban canyon environments.
Impact:
Operations in urban canyons and interior environments are among the most dangerous activities in which our sponsors are currently engaged. The 3D sensing and exploitation provided by simultaneous localization and mapping will greatly enhance the utility of small unmanned vehicles to provide situational awareness in settings not well served by currently deployed systems.
Approved for Public Release: 06-1320 Presentation [PDF]
A Multi-Source Recommender System for Intelligence Analysis Tom Bartee, Principal Investigator Problems:
Intelligence analysts have an enormous amount of information available to support their investigations. Unfortunately, analyst-driven information searches may miss important information. Searches are time-consuming, and the information and supporting applications change constantly. Furthermore, analysts are prone to human judgment biases that can lead them to miss or discount information that does not directly support their hypotheses.
Objectives:
We will develop a Recommender System that supports intelligence analysis through the automatic retrieval of information that is relevant to the specific current needs of the analyst. This information will save analysts time and facilitate unbiased decision making by providing a range of data that would both support and refute analysts' working hypotheses.
Activities:
We will implement the core infrastructure for instrumentation of analysts' activities, identification of key events, and browsing of information retrieved by the system. Next we will develop an Expert Task Model for our core challenge problem and implement Event Processor scripts to automate searches. Finally, we will move our system into an analytic environment and perform a formative evaluation.
Impact:
Our approach for a Task-Oriented Recommender System will be applicable to a broad range of intelligence problems, as well as to problems outside the intelligence community. The generic architecture supports software reuse across domains.
Approved for Public Release: 05-1252 Presentation [PDF]
Advanced Perception and Unmanned Ground Vehicles Bob Grabowski, Principal Investigator Problems:
Perception is essential for competent ground robots. Robots derive perception of the world directly through sensors. Operational expectations for existing robots far exceed the capabilities of current sensor technology, as evidenced by recent events such as the DARPA Grand Challenge. As the mission space becomes more complex, sensor technology must keep pace.
Objectives:
We will demonstrate improved perception capability on an unmanned ground vehicle (UGV) and provide a framework for defining the necessary sensing capabilities for UGVs. We will demonstrate how to enhance vehicle perception by combining multiple sensors to overcome the weakness of individual technologies, and investigate novel techniques such as sensor fusion, sensor cueing, sensor steering, and mosaicing.
Activities:
We will identify capabilities of existing sensor technology with regard to UGV maneuvering, navigation, safety, and awareness, and characterize the strengths and weaknesses of each approach. We will develop new methods for combining and fusing information from multiple sensors to overcome identified limitations and gaps, and then test and verify new approaches in a representative urban environment on a real UGV.
Impact:
This research will shape future approaches to development of sensing capabilities for UGVs -- both the vehicles and the humans who must operate around them. By developing and testing the results on an existing autonomous vehicle, this research will also yield a vehicle capable of conducting more complex missions in realistic, operationally relevant environments.
Approved for Public Release: 06-1432 Presentation [PDF]
Cooperative Robot Pairing Bob Grabowski, Principal Investigator Problems:
The military services and the DoD are counting heavily on future robotics capabilities. The current paradigm focuses on the development and execution of a single robot platform type. Mission effectiveness is constrained by the size, processing, and sensing of that platform. These limitations may unnecessarily limit the range of missions that unmanned systems can support.
Objectives:
Our research will combine robots across diverse scales and dimensions, demonstrating enhanced robot functionality over that of a single robot platform. We will show how low-level autonomy can significantly reduce dependencies on human operators. We will combine the strengths of a large autonomous truck (Meteor), a small ground vehicle (PackBot), and an autonomous escort helicopter for application to military missions.
Activities:
We will continue the full integration of display and control of the Meteor and PackBot via the Command Post of the Future, and implement an autonomous docking behavior for the PackBot and the Meteor. We will also investigate the nature of perimeter patrol and security as a mission for cooperative robots, and include a rotary-wing UAV escort.
Impact:
We will demonstrate the increased utility of cooperating robots for specific application to the patrol mission. We will develop individual robotic functionalities that can transition rapidly to the military community. Research will lead to fielded platforms useful for future testing of unmanned systems relevant to MITRE sponsors.
Approved for Public Release: 06-0057 Presentation [PDF]
Historical Compliance Behavior Predictive Model Dave DeBarr, Principal Investigator Problems:
The tax gap -- the difference between taxes that were collected and taxes that should have been collected -- is estimated to be over $300 billion per year. The tax gap comprises non-filing (of tax returns), underpayment (of taxes), and underreporting (of tax liability). Underreporting, due to understated income or overstated deductions, makes up over 80 percent of the tax gap.
Objectives:
The IRS uses computer-based screening of tax returns to identify compliance issues. Some of these algorithms are based on heuristics, while others are based on models trained using audit data. The goal of this project is to determine if using the characteristics of a taxpayer's compliance history improves the performance of one or more of these targeting algorithms.
Activities:
We will work with staff from the IRS Research, Analysis, and Statistics division to select one or more targeting models, establish criteria for measuring performance, and explore the available data. For selected applications, we will iteratively preprocess the data, construct new models, and evaluate these models. We will write up results during the final quarter of FY07.
Impact:
Our ultimate goal is to improve the performance of computer-based models used for targeting compliance issues at the IRS. We hope that considering the characteristics of compliance history will result in fewer false positives, fewer false negatives, and an overall increase in the amount of revenue collected per hour of staff effort.
Approved for Public Release: 07-0460
Image Regularization for Biometric Face ID Jeff Colombe, Principal Investigator Presentation [PDF]
Inference Rules for Joint Mission Assessment Lewis Loren, Principal Investigator Problems:
Planning and execution disconnections across strike and ISR missions detrimentally impact the Air Force's responsiveness. Currently, there is no efficient means of determining ATO mission status or an aircraft's munitions load and, while the target list is constantly evolving, such changes frequently fail to find their way to the ISR deck. Such shortfalls prevent a real-time assessment of combat objectives.
Objectives:
Prototypes will be developed to address planning and execution shortfalls by making inferences regarding mission status and combat objectives. Such inference rules will also improve situation awareness for operators controlling ISR assets, strike assets, and logistics support, and will support effects-based operations by enabling a mapping between real-time activities and the objectives and timetable.
Activities:
We will develop prototypes with interfaces to mission planning systems. These prototypes will make inferences regarding mission status and combat objectives. A beta version of the prototypes will be ready by the second quarter, and will be presented to operators for feedback and assessment by the third quarter.
Impact:
The developed prototypes will permit MITRE and ESC to respond quickly to pervasive problems of great importance to operators. We will demonstrate the value of inference rules and their applicability to time-sensitive targeting, ISR management, and effects-based operations. Stakeholders include MIT, the C2 Battlelab, and mission planning systems.
Approved for Public Release: 05-1394 Presentation [PDF]
ISR Forensics Curtis Brown, Principal Investigator Problems:
Forensic analysis will become one of the future drivers of both exploitation tool development and multi-intelligence (multi-INT) data archives. To be successful in exploiting the ISR data in large multi-INT databases, forensics analysts will need to combine ground-moving target indicator data with data from other sources of intelligence and will require easy-to-use tools to access, navigate, and process it.
Objectives:
Our objective is to create visualization tools to support the forensic analysis of ISR data. These tools will provide a framework for exploring automation through the incorporation of advanced tracking technology.
Activities:
We will develop a set of tools that exploit geospatial and temporal data, and then apply the tools to synthetic ground truth data sets that we will generate. Advanced tracking technology, including Multiple Hypothesis Tracker traceback, will be developed and integrated to aid the analyst. We will document lessons learned and demonstrate the forensic environment at the Technology Symposium.
Impact:
We will make our exploitation tools available to analysts. Lessons learned will be fed back both to multi-INT database efforts as well as exploitation tool development efforts.
Approved for Public Release: 05-0200 Presentation [PDF]
Multimodal Medical Data Capture & Representation Qian Hu, Principal Investigator Problems:
An obstacle to widespread adoption of Electronic Medical Record (EMR) systems is the difficulty of capturing structured clinical information from unstructured data. Automatic speech recognition (ASR) and handwriting recognition (HR) have been applied to EMR systems without much success, due to lack of required accuracy, poor integration with hospitals' workflow, and the problem of converting doctors' natural speech or handwritten notes to a standard format.
Objectives:
This project will research and develop algorithms and methodologies to (1) enhance and adapt ASR/HR to the medical domain/specialty to enable speech and handwriting as reliable data capture methods, (2) use a multimodal method to interface with EMR systems, and (3) use information from multimodalities, medical domain knowledge, and prior EMRs to interpret and convert captured free speech and handwriting to standard expressions.
Activities:
We will evaluate medical ASR engines and HR to identify areas for enhancement and adaptation, evaluate open source EMR systems, interview doctors, and visit hospitals to understand the workflow and EMR requirements. We will establish baseline performance using existing technology and build training corpora for ASR/HR, and then design a multimodal research prototype.
Impact:
Multimodal technology and methodologies for medical data capture and representation can revolutionize the creation of EMR. Going beyond verbatim conversion to encode conceptual underpinnings is critical for interoperability among healthcare and research institutes. We have shared our evaluation methodology and some results of ASR/HR with non-profit health organizations and have been invited to consult on their design and evaluation of EMR systems.
Approved for Public Release: 07-0268 Presentation [PDF]
Structured ISR Fusion Walter Kuklinski, Principal Investigator Problems:
The importance of developing fusion algorithms to improve the warfighting capability derived from Intelligence Surveillance and Reconnaissance (ISR) systems is well documented. Many operational requirements mandate levels of performance that can only be obtained by fusing data from multiple sensors and platforms. However, only limited attempts have been made to develop formal design procedures for data fusion systems.
Objectives:
The objective of this MSR is to use formal mathematical methods such as Random Set Theory and Category Theory as frameworks to design and implement fusion algorithms and to predict fusion system performance. We will develop and implement a widely applicable methodology for fusion system design and analysis.
Activities:
We will develop and evaluate, using both simulated and field data, multi-target tracking algorithms that fuse conventional radar SMTI data with additional data sources including SIGINT, COMINT and ELINT, using both Random Set Theory and Category Theory frameworks.
Impact:
The successful completion of this MSR will both provide a significant advance in the general area of fusion system analysis and design, and yield specific fusion algorithms applicable to the future fusion requirements of many MITRE sponsors.
Approved for Public Release: 05-0178 Presentation [PDF]
^TOP | 
