Advances in pervasive communication technology have enabled many unconventional applications that facilitate and improve the safety and quality of life in modern society. Among emerging applications is situational awareness where individuals and first-responders receive timely alerts about serious events that could have caused the interruption of the services provided by the communication infrastructure such as cellular networks, Wi-Fi hotspots, etc. Another example is when exchanging road conditions between peer-to-peer networked vehicles without the involvement of roadside units. The popularity of smart portable devices such as iPhone and Android powered phones and tablets has made them an attractive choice that can play a role in the realization of these emerging applications. These devices support multiple communication standards and thus enable Device-to-Device (D2D) data exchange at an increased level of convenience. By using technologies such as Bluetooth, Wi-Fi ad-hoc mode and Wi-Fi Direct, these devices are able to communicate without the need for any communication infrastructures. In addition, many of these devices are equipped with sensors that can provide a wealth of information about the surroundings once their readings are aggregated.
However, most existing protocols for data sharing among devices either require an internet connection, which may not be available and may incur extra costs in some cases, or suffer from the device’s operating system limitations. Actually there is no existing solution that allows a set of devices to start sharing data dynamically without forcing the users to apply an elaborate procedure for setting up a group. These shortcomings render existing solutions unsuitable for emergency cases. In this dissertation proposal, we tackle such a problem by developing a framework for enabling data exchange in a cost-effective and timely manner through the establishment of peer-to-peer links among smart devices. In addition, our framework opts to minimize the user required interaction for setting up a connection and overcome the limitations of the operating system.
Our framework consists of a set of protocols for group data exchange using Wi-Fi Direct on Android devices. First we present an Efficient and Lightweight protocol for peer-to-peer Networking of Android smart devices over Wi-Fi Direct (ELN). ELN main goal is to overcome the Wi-Fi Direct support limitations in Android, thus allowing the devices in one Wi-Fi Direct group to communicate together. The ELN protocol is validated by implementing a group chatting application. In addition, we present a protocol for Alert Dissemination using Service discovery (ADS) in Wi-Fi Direct. ADS uses the service discovery feature of Wi-Fi Direct for distributing alerts to nearby devices without requiring any prior connections and thus avoids the setup delay in creating Wi-Fi Direct groups and the limitations of multi-group connectivity in Android. ADS is validated by implementing a hazard propagation application for Android. Finally, we present an Efficient Multi-group formation and Communication (EMC) protocol for Wi-Fi Direct. EMC exploits the battery specifications of the devices to qualify potential group owners and enable dynamic formation of efficient groups. Moreover, EMC allows data exchange between different Wi-Fi direct groups. Part of our implementation of EMC in Android involves the modification of the Android source code to allow multi-group support. A chat application is developed to validate EMC.
To complete the dissertation, we plan to extend EMC by replacing the static assignment of devices’ addresses in our current implementation with an IP address negotiation protocol that runs before creating groups. Such an extension would give greater flexibility in adapting EMC. In addition, we plan to define some criteria for selecting proxy members in order to allow maximum coverage and allow the D2D communication to span a larger geographical area. In addition, we will develop a simulator to do large scale testing for the proposed framework. Finally, we would like to explore the use of dual transceivers in order to increase the robustness of D2D connections when the wireless channels are subject to varying level of interference; particularly we like to investigate the integration of Bluetooth Low Energy within our framework to enable group membership of nodes that do not have Wi-Fi Direct or suffer interference that makes the Wi-Fi Direct links unstable.
Committee: Drs. Mohamed Younis (Chair), Charles Nicholas, Chintan Patel, Tinoosh Mohsenin
Motivated by such learning in nature, the problem of Learning from Demonstration is targeted at learning to perform tasks based on observed examples. One of the approaches to Learning from Demonstration is Inverse Reinforcement Learning, in which actions are observed to infer rewards. This work combines a feature based state evaluation approach to Inverse Reinforcement Learning with neuroevolution, a paradigm for modifying neural networks based on their performance on a given task. Neural networks are used to learn from a demonstrated expert policy and are evolved to generate a policy similar to the demonstration. The algorithm is discussed and evaluated against competitive feature-based Inverse Reinforcement Learning approaches. At the cost of execution time, neural networks allow for non-linear combinations of features in state evaluations. These valuations may correspond to state value or state reward. This results in better correspondence to observed examples as opposed to using linear combinations.
This work also extends existing work on Bayesian Non-Parametric Feature construction for Inverse Reinforcement Learning by using non-linear combinations of intermediate data to improve performance. The algorithm is observed to be specifically suitable for a linearly solvable non-deterministic Markov Decision Processes in which multiple rewards are sparsely scattered in state space. Performance of the algorithm is shown to be limited by parameters used, implying adjustable capability. A conclusive performance hierarchy between evaluated algorithms is constructed.
Committee: Drs. Tim Oates, Cynthia Matuszek and Tim Finin
Computing solutions to intractable planning problems is particularly problematic in dynamic, real-time domains. For example, visitation planning problems, such as a delivery truck that must deliver packages to various locations, can be mapped to a Traveling Salesman Problem (TSP). The TSP is an NP-complete problem, requiring planners to use heuristics to find solutions to any significantly large problem instance, and can require a lengthy amount of time. Planners that solve the dynamic variant, the Dynamic Traveling Salesman Problem (DTSP), calculate an efficient route to visit a set of potentially changing locations. When a new location becomes known, DTSP planners typically use heuristics to add the new locations to the previously computed route. Depending on the placement and quantity of these new locations, the efficiency of this adapted, approximated solution can vary significantly. Solving a DTSP in real time thus requires choosing between a TSP planner, which produces a relatively good but slowly generated solution, and a DTSP planner, which produces a less optimal solution relatively quickly.
Instead of quickly generating approximate solutions or slowly generating better solutions at runtime, this dissertation introduces an alternate approach of precomputing a library of high-quality solutions prior to runtime. One could imagine a library containing a high-quality solution for every potential problem instance consisting of potential new locations, but this approach obviously does not scale with increasing problem complexity. Because complex domains preclude creating a comprehensive library, I instead choose a subset of all possible plans to include. Strategic plan selection will ensure that the library contains appropriate plans for future scenarios.
Committee: Drs. Marie desJardins (co-chair), Tim Finin (co-chair), Tim Oates, Donald Miner, R. Scott Cost
CSEE professor Cynthia Matuszek will teach a new special topics course this spring on Principles of Human-Robot Interaction. The graduate level course (CMSC 691-08) will meet on Tuesday and Thursdays from 4:00 to 5:30pm in 013 Sherman Hall.
Robots are becoming ubiquitous. From Roombas in our homes, to surgical robots in hospitals, to giant manipulators that assemble cars, robots are everywhere. In the past, robots have only ever interacted with highly trained experts. Now, as they are being deployed more widely, we must address new questions about how our robots can interact day-to-day with end users — non-experts — safely, usefully, and pleasantly. This new area of research is called Human-Robot Interaction, or HRI.
This 3-credit special topics course aims to introduce students to current research in HRI and provide hands-on experience with HRI research. Students will explore the diverse range of research topics in this area, learn to identify HRI problems in their own research, and carry out a collaborative project involving human-robot interactions. Topics to be covered include:
Students may benefit from having some previous coursework or experience in AI, machine learning, or robotics, but none are necessary. Undergraduate students can enroll with the instructor’s permission. For more information, contact Dr. Matuszek at cmat at umbc.edu.
CRA-W is now accepting applications for Grad Cohort 2016, a two-day workshop during which participants will learn graduate school survival skills, receive mentoring, and develop networks with senior female computing researchers. This is a great opportunity for female graduate students to build mentoring relationships and develop peer networks to form the foundation of their graduate career and beyond.p
Female graduate students in their first three years are eligible to apply. Reasonable travel expenses, meals, and lodging will be provided for students chosen to participate in this program.
The Grad Cohort 2016 workshop will be held at the Hilton San Diego Bayfront in San Diego, California, on April 15-16, 2016. The application deadline is 30 November 2015. Apply online here and get more information at the Grad Cohort 2016 Workshop site.
Currently, software and hardware system components are trending toward modularized and virtualized as atomic services on the cloud. A number of cloud platforms or marketplaces are available where everybody can provide their system components as services. In this situation, service composition is essential, because the functionalities offered by a single atomic service might not satisfy users’ complex requirements. Since there are already a number of available services and significant increase in the number of new services over time, manual service composition is impractical.
In our research, we propose computer-aided methods to help find and compose appropriate services to fulfill users’ requirement in large scale service network. For this purpose, we explore the following methods. First, we develop a method for formally representing a service in term of composability by considering various functional and non-functional characteristics of services. Second, we develop a method for aiding the development of the reference ontologies that are crucial for representing a service. We explore a bottom-up-based statistical method for the ontology development. Third, we architect a framework that encompasses the reference models, effective strategy, and necessary procedures for the services search and composition. Finally, we develop a graph-based algorithm that is highly specialized for services search and composition. Experimental comparative performance analysis against existing automatic services composition methods is also provided.
Commitee: Drs. Yun Peng (chair), Tim Finin, Yelena Yesha, Milton Halem, Nenad Ivezic (NIST) and Boonserm Kulvatunyou (NIST)
Objects in the world around us are made of a myriad of materials, both metallic and non-metallic. Most non-metallic materials scatter light in varying amounts within their surfaces, giving softer, more saturated diffuse colors and softer-edged shadows. This effect, subsurface scattering, is important to make translucent objects look realistic. Non-metallic objects that are opaque also scatter light, just at a very small distance. These non-metallic materials may look somewhat translucent at very close viewing distances, but from farther away they exhibit a more opaque, but still soft diffuse appearance. To shade these objects realistically from all distances, a method is needed to model subsurface scattering effects at close ranges and to smoothly transition to a soft diffuse reflection at larger viewing distances. We present a method that takes advantage of graphics processor texture filtering hardware to linearly filter maps that encode diffuse reflection and translucency information and to interpolate between a close-range subsurface scattering effect and a long-range reflectance function.
Committee: Drs. Marc Olano (Advisor, Chair), Penny Rheingans, Jian Chen
Giving talks is one of the core tasks of a researcher. Technical presentations are how we accomplish some of our most important tasks: talks are the first step in getting other people excited about our work, getting suggestions and feedback, teaching, and applying for jobs and grants. Nonetheless, the art and science of giving a really good technical talk is one we are more likely to leave to chance than to deliberately train in. Not only does this mean we aren’t accomplishing everything we could with our presentations; we’re missing a chance to distinguish ourselves by improving a comparatively rare — but learnable — skill.
In this talk, I will describe the idea of the “culture of conveying information,” and give a number of specific suggestions for improving technical talks — including tools, rules of thumb, social conventions, and suggestions for making your talks engaging, informative, and memorable.
Cynthia Matuszek is an Assistant Professor at the University of Maryland, Baltimore County’s Computer Science and Electrical Engineering department where she heads the Interactive Robotics and Language lab. She completed her Ph.D. at the University of Washington in 2014, where she was a member of both the Robotics and State Estimation lab and the Language, Interaction, and Learning group. She is published in the areas of artificial intelligence, robotics, ubiquitous computing, and human-robot interaction. Her research interests include human-robot interaction, natural language processing, and machine learning.
Due to the increase in security concerns, automated detection of abandoned objects has become an important application in video surveillance. Because of its increasing importance, a number of techniques have been proposed recently to automatically detect abandoned objects. The general procedure implemented for detection of abandoned objects includes background subtraction or foreground object extraction followed by post-processing steps in order to classify the foreground object as an abandoned or non-abandoned object. However, these techniques make use of a number of user-defined parameters such as track time, co-ordinates of the object/owner, the vicinity of the object, and properties of the object such as its shape, color, among others.
In this thesis, we present a new technique based on blind source separation (BSS) for detection of abandoned objects that does not keep track of the extracted objects or owners and does not require a dual background scheme for stationary object extraction. Order selection is an important step for our implementation of blind source separation based scheme since this step captures the signals with high energy and disregards signals that are not relevant to the detection of abandoned objects. In this thesis, we show that the performance of ICA improves when an algorithm that assumes a flexible source distribution along with multiple types of diversity, such as higher-order statistics and sample dependence is used for the estimation of the source components. ICA, however, can only model one dataset at a time, thus limiting its usage to monochrome frames. In order to address this issue, we also present another implementation of blind source separation called independent vector analysis (IVA), a recent extension of ICA to multiple data that takes the dependence across multiple datasets into account while retaining the model of independent components within each dataset. We show that the proposed blind source separation techniques performs successfully in complicated scenarios such as crowd, occlusion, and illumination changes.
Committee: Drs. Tulay Adali (chair), Joel Morris and Mohamed Younis