Cooperative Localization

The problem of cooperative localization entails answering the following question, “Given a set of N points in a k-dimensional metric space and their corresponding pairwise distance/proximity measurements, can we find a conformation or relative positions of the points that satisfy these pairwise measurements?”. The problem is of significant interest in varied application areas such as sensor networks, transportation systems, multimedia processing, robotics, biology, non-linear dimensionality reduction etc. My current research is largely focused on the theoretical, algorithmic and practical aspects of this problem.

Collaborative High Accuracy Localization for Intelligent Transportation Systems

High accuracy localization for precise localization of vehicles in multipath environments is of significant interest for critical safety applications. Our research focuses on robust peer-to-peer localization algorithms in multipath environments under bandwidth constraints. The U.S Department of Transportation recently recognized our work by selecting us as one of the six winners of the U.S. DoT Connected Vehicle Technology Challenge. Relevant references are conferences papers [4], [5], [6], [7] and technical reports [1], [2]. This is joint work with Prof. Kannan Ramchandran and Prof. Raja Sengupta.

Multimodal Location Estimation for consumer generated multimedia content

Associating geo-locations with consumer produced multimedia content such as videos and photos publicly available on Flickr, Picasa etc, is a problem of significant recent research interest. Applications range from location based services to large scale 3D modeling. Our research work explores smart graphical model based inference algorithms for this problem of geotagging using multimodal features such as audio, video and textual tags. Relevant reference is our recent conference paper [2]. This is joint work with Jaeyoung Choi, Dr. Gerald Friedland and Prof. Kannan Ramchandran.

High Accuracy 3D Localization in Mobile Multipath Environments using RFID Supertags

High-accuracy indoor positioning is mandated in many applications such as robotic positioning and navigation, first responder services etc. The non-availability of GPS requires us to develop augmenting sensing technologies to enable positioning. We propose a novel architecture for precise 3D localization in indoor multipath environments using RFID “supertags” that enables a robust inexpensive self-calibrating navigation system.             

          This is joint work with Prof. Kannan Ramchandran.

Convex Relaxations for Localization

Estimating the point conformation given a subset of the pairwise distance measurements is NP-hard. The problem is further interesting when a fraction of these measurements are corrupted by outliers. We explore SDP relaxation and matrix completion techniques and provide sufficient conditions under which the problem can be solved. The problem is of significant interest in applications like protein molecular conformation, manifold learning, sensor node localization etc. Relevant reference is our recent conference submission [1]. This is joint work with Prof. Kannan Ramchandran.

Behavioral Economics and Sustainable Transportation

Sustainable transportation is one of the major challenges of the 21st century. The success of achieving sustainable transportation largely depends on individual willingness to adopt efficient modes of transportation and change their existing travel pattern and mode choices. This necessitates obtaining utility models of individuals in order to design effective persuasive strategies tailored to that individual. We explore techniques inspired from the area of behavioral economics to design experiments from which we can estimate individual utility functions in a statistically significant and reliable manner. We further develop persuasive technology platforms to induce behavior change and study the effectiveness of our strategies.

The Quantified Traveler

Can we use GPS tracking data to analyze a user’s mode choice? Does providing information such as emissions, cost, calories and time for a user’s travel affect his travel behavior? Does social comparison help? These are some of the questions we ask as part of this project. The project is largely spearheaded by Prof Raja Sengupta and Prof Joan Walker and their graduate students. Relevant reference is conference paper [3].

Hop-count based Self-Localization for Sensor Networks.

Estimating positions of nodes in a sensor network is a problem of significant interest. These systems are challenged by energy and communication constraints. Our research focused on developing a theoretical relationship between hop-counts and Euclidean distances and we explored graphical model based algorithms for localization using hop-counts. Relevant reference is journal paper [1]. This is joint work with Swaprava Nath, Prof. P. Vijay Kumar and Prof. Anurag Kumar.

Detecting an intruder in a sensor field with minimal communication between sensors is a topic of interest in sensor networks specifically for military applications. We focused on obtaining an optimal tradeoff between the detection performance and the energy spent in detection. We were able to make an interesting connection between the problem of detection and communication complexity theory that helped obtain bounds on the energy trade-off.  We received the IEEE DCOSS'08 Best paper award (Algorithms track), for this work. Relevant references are conference papers [8], [9] and journal paper [2]. This is joint work with Tarun Agarwal and Prof. P. Vijay Kumar.

Distributed detection in sensor networks

Image resizing and its applications

Given the variety of electronic devices with different screen resolutions, it is necessary to develop efficient image resizing algorithms. We propose an efficient resizing algorithm in the image domain and evaluate its performance on an Analog Devices video processor. Relevant reference is conference paper [9].