Defense Notices

In a multi-agent system, an idle agent may be available to assist other agents in the system. An agent architecture called intent recognition is proposed to accomplish this with minimal communication. 
In order to assist other agents in the system, an agent performing recognition observes the tasks other agents are performing. Unlike the much studied field of plan recognition, the overall intent of an agent is recognized instead of a specific plan. The observing agent may use capabilities that it has not observed. This study focuses on the key research questions of: (1) What are intent recognition systems? (2) How can these be used in order to have agents autonomously assist each other effectively and efficiently? A conceptual framework is proposed for intent recognition systems. An implementation of the conceptual framework is tested and evaluated. We hypothesize that using intent recognition in a multi-agent system increases utility (where utility is domain specific) and decreases the amount of communication. We test our hypotheses using two experimental series in the domains of Box Pushing, where agents attempt to push boxes to specified locations; and Cow Herding, where agents attempt to herd cow agents into team corrals. A set of metrics, including task time and number of communications, is used to compare the performance of plan recognition and intent recognition. In both sets of experimental series, intent recognition agents communicate fewer times than plan recognition agents. In addition, unlike plan recognition, when agents use the novel approach of intent recognition, they select unobserved actions to perform, which was seen in both experimental series. Intent recognition agents were also able to outperform plan recognition agents by sometimes reducing task completion time in the Box Pushing domain and consistently scoring more points in the Cow Herding domain.

We examine the problem of determining a decision threshold for the binary hypothesis test that naturally arises when a radar system must decide if there is a target present in a range cell under test. Modern radar systems require predictable, low, constant rates of false alarm (i.e. when unwanted noise and clutter returns are mistaken for a target). Measured clutter returns have often been fitted to heavy tailed, non-Gaussian distributions. The heavy tails on these distributions cause an unacceptable rise in the number of false alarms. We use the class of spherically invariant random vectors (SIRVs) to model clutter returns. SIRVs arise from a phenomenological consideration of the radar sensing problem, and include both the Gaussian distribution and most commonly reported non-Gaussian clutter distributions (e.g. K distribution, Weibull distribution). We propose an extension of a prior technique called the Ozturk algorithm. The Ozturk algorithm generates a graphical library of points corresponding to known SIRV distributions. These points are generated from linked vectors whose magnitude is derived from the order statistics of the SIRV distributions. Measured data is then compared to the library and a distribution is chosen that best approximates the measured data. Our extension introduces a framework of weighting functions and adaptively scaling of the measured data. Further, we extend the Ozturk algorithm to both a distribution classi fication technique as well as a method of determining an adaptive threshold in data that may not belong to a known distribution. Special attention is paid to producing a robust, adaptive estimation of the detection threshold.

Communication networks, in particular the Internet, face a variety of challenges that can disrupt our daily lives resulting in the loss of human lives and significant financial costs in the worst cases. We define challenges as external events that trigger faults that eventually result in service failures. Understanding these challenges accordingly is essential for improvement of the current networks and for designing Future Internet architectures. This dissertation presents a taxonomy of challenges that can help evaluate design choices for the current and Future Internet. Graph models to analyse critical infrastructures are examined and a multilevel graph model is developed to study interdependencies between different networks. Furthermore, graph-theoretic heuristic optimisation algorithms are developed. These heuristic algorithms add links to increase the resilience of networks in the least costly manner and they are computationally less expensive than an exhaustive search algorithm. The performance of networks under random failures, targeted attacks, and correlated area-based challenges are evaluated by the challenge simulation module that we developed. The GpENI Future Internet testbed is used to conduct experiments to evaluate the performance of the heuristic algorithms developed.

The surface-elevation of ice sheets and sea ice is currently measured using both satellite and airborne radar altimeters. These measurements are used for generating mass balance estimates of ice sheets and thickness estimates of sea ice. However, due to the penetration of the altimeter signal into the snow there is ambiguity between the surface tracking point and the actual surface location which produces errors in the surface elevation measurement. Until now there is no comprehensive study done to address how the penetration of the Ku-band signal affects the shape of the return signal over various snow zones and sea ice. Therefore, it is important to study the effect sub-surface scattering and seasonal variations in the properties of snow pack have on the return waveform to correctly interpret the satellite radar altimeter data. To address this problem, an ultra-wide bandwidth Ku-band radar altimeter was developed at the Center for Remote Sensing of Ice Sheets (CReSIS). The CReSIS Ku-band Altimeter (CKA) operates over the frequency range of 12 to 18 GHz providing very fine resolution to resolve the sub-surface features of the snow. The CKA is design to encompass the frequency band of SIRAL, a satellite radar altimeter on board CryoSat-2, operating from 13.4 to 13.75 GHz. The data from CKA can be used to simulate SIRAL data, and the simulated SIRAL waveforms can help us understand the effect of signal penetration and sub-surface scattering on the low bandwidth satellite altimeter. The extensive CKA data collected as a part of the Operation Ice Bridge (OIB) campaign can be used to interpret SIRAL data over surfaces with varying snow conditions. The goal of this research is to use modeling and data inter-comparisons from CKA and satellite measurements to investigate the effect of signal penetration into snow and geophysical snow conditions on the retrieval of surface elevation from satellite radar altimeters, such as SIRAL. Based on the results of this investigation, the plan is to improve the tracking algorithms used by SIRAL to effectively track the actual surface location accurately.

As well recognized, healthcare information is growing exponentially and is made more available to public. Frequent users such as medical professionals and patients are highly dependent on the web sources to get the appropriate information promptly. However, the trustworthiness of the web information can be hardly discriminated due to the fast and augmentative properties of the Internet. Most search engines provide relevant pages to given keywords, but the results might contain unreliable or biased information. 

In this dissertation, I proposed a new system named HealthTrust, which automatically assesses the trustworthiness of healthcare information over the Internet. First, in the first phase, a new ranking algorithm for structure-based analysis is adopted. The basic hypothesis is that trustworthy pages are more likely to link to trustworthy pages. In this way, the original set of positive and negative seeds will propagate over the Web graph. Next, in the second phase, the content-consistency between general healthcare-related webpages and trusted sites is evaluated using information retrieval techniques to evaluate the context of the webpage. In addition, sentence modeling is employed to generate contents-based ranking for each page. Finally, an iterative algorithm is developed to integrate the two components.

Existing RF path loss models are useful for prediction but do not necessarily exploit additional knowledge such as differing terrain features. This research will examine the relationship between terrain type (determined by public GIS data sets) and empirical path loss model parameters through the use of a large scale data collection platform. A large scale measurement campaign will be undertaken to sample the UHF DTV bands from a variety of environments using a set of portable software defined radio sensors. Machine learning and geostatistical algorithms will then be used to learn path loss parameters for generalized terrain types.

"Assistive robotics for the elderly has become a significant area of research, driven primarily by a rapidly aging global population. Between 2011 and 2050, the number of people aged 60 and over is expected to climb from 893 million to 2.4 billion. In addition to a rapidly aging global population, a growing shortage of caretakers has placed additional urgency on the search for alternative solutions. 

Despite the potential benefits of assistive robotics, one significant hurdle that remains is designing robots that the elderly are willing to use. Not only must assistive robots be effective at monitoring and caring for the elderly, but they must also be acceptable to a wide range of elderly individuals. While a variety of factors can influence the acceptability of a robot, past research has focused primarily on the physical embodiment of the robot. 

Social robotics, however, uses human-robot interactions to study the many social factors that can influence the acceptability of a robot, including affective behaviors, or behaviors that simulate personality and emotion. While the majority of research in affective behaviors has focused on facial expressions, these methods require sophisticated anthropomorphic representations, which are not generally preferred by the elderly. 

However, in addition to facial expressions, body language can also be an effective communicator of emotions and personalities. This research will demonstrate the effectiveness of using non-verbal behaviors to simulate a variety of personalities and emotions on the Aldebaran Nao, as well as the impact these behaviors have on the elderly’s trust and acceptance of the assistive robot. By adapting the personalities and emotions of an assistive robot both to the task it is performing, as well as to individual elderly users, this research will ultimately enable assistive robots to perform as more effective caretakers for the elderly."

Software development has been and continues to be a difficult enterprise. As early as the 1960’s, computer experts recognized the need to develop working and reliable software, within budget and on time. One of the major obstacles to successful software development is the inevitability of changing requirements. Agile software development methods, such as Extreme Programming and Scrum, emerged in the 1990’s as responses to deal with constant change. However, agile software development methods have their own set of weaknesses. Two specific weaknesses with Scrum are a lack of initial planning and a lack of an overall architecture. 
Military operations are another field that must deal with constantly changing requirements in complex environments. In response to this inescapable change, the military these days has primarily employed mission command to direct operations. Mission command is the philosophy where a commander gives subordinates his/her intent and desired end state for an operation, and then the subordinates have appropriate flexibility to operate to achieve that intent and end state. This research effort seeks to use inspirations from mission command to improve certain aspects of agile software development, namely Scrum, both in terms of the process and the product. 
This research effort seeks to address the lack of initial planning in Scrum with the addition of a Product Design Meeting at the onset of the process. This effort also addresses the lack of an overall architecture using two artifacts derived from mission command, the product’s end state and lines of effort (LOEs) necessary to achieve that end state. The addition of the Product Design Meeting, product end state, and LOEs will add more formalism to an agile method. However, we hypothesize that the benefits of adding those techniques will offset any perceived loss of agility.

One of the key requirements for deriving more realistic ice sheet models is to obtain a good set of basal measurements that enable accurate estimation of bed roughness and conditions. For this purpose, 3D tomography of the ice bed has been successfully implemented with the help of angle of arrival estimation (AoA) algorithms such as multiple signal classification (MUSIC) and maximum likelihood estimation (MLE) techniques. These methods have enabled fine resolution in the cross-track dimension using synthetic aperture radar (SAR) images obtained from single pass multichannel data. This project analyzes and compares the results obtained from the spectral MUSIC algorithm, an alternating projection (AP) based MLE technique, and a relatively recent approach called the reiterative superresolution (RISR) algorithm. While the MUSIC algorithm is more attractive computationally compared to MLE, the performance of the latter is known to be superior in a low signal to noise ratio regime. The RISR algorithm takes a completely different approach by using a recursive implementation of the minimum mean square error (MMSE) estimation technique instead of using the sample covariance matrix (SCM) that is central to subspace based algorithms. This renders the algorithm more robust in scenarios where there is a very low sample support. The SAR focused datasets provide a good case study to explore the performance of the three techniques to the application of ice sheet bed elevation estimation.

Synchronization is a critical operation in digital communication systems, which establishes and maintains an operational link between transmitter and the receiver. As the advancement of digital modulation and coding schemes continues, the synchronization task becomes more and more challenging since the new standards require high-throughput functionality at low signal-to-noise ratios (SNRs). In this work, we address feedforward synchronization of continuous phase modulations (CPMs) using data-aided (DA) methods, which are best suited for burst-mode communications. In our transmission model, a known training sequence is appended to the beginning of each burst, which is then affected by additive white Gaussian noise (AWGN), and unknown frequency, phase, and timing offsets. 

Based on our transmission model, we derive the optimum training sequence for DA synchronization of CPM signals using the Cramer-Rao bound (CRB), which is a lower bound on the estimation error variance. It is shown that the proposed sequence minimizes the CRB for all three synchronization parameters, and can be applied to the entire CPM family. We take advantage of the structure of the optimized training sequence in order to derive a maximum likelihood joint timing and carrier recovery algorithm. Moreover, a frame synchronization algorithm is proposed, and hence, a complete synchronization scheme is presented in this work. 

The proposed training sequence and synchronization algorithm are extended to shaped-offset quadrature phase-shift keying (SOQPSK) modulation, which is considered for next generation aeronautical telemetry systems. Here, it is shown that the optimized training sequence outperforms the one that is defined in the draft telemetry standard as long as estimation error variances are considered. The overall bit error rate suggest that the optimized training sequence with a shorter length can be utilized such that the SNR loss is less than 0.5 dB of an ideal synchronization scenario.