11th Undergraduate Research Symposium (2017)
Held 21 April 2017 in 011 Sitterson Hall
Schedule
Click on a presentation title to read the abstract and presenter bio or watch the presentation on YouTube.
3:30 – Session 1
“Synthetic Data for Crowd and Human Understanding” by Anson Wong, supervised by Dinesh Manocha
“Evaluation of the Performance and Cost of Cloud-Based Robot Motion Planning” by Jonathan Lynn, supervised by Ron Alterovitz
“Graph-Based Patrol Techniques for Use in Video Games” by Ryan Gibson, supervised by Diane Pozefsky
4:15 – Session 2
“Variance Reduction in Global Illumination with Monte Carlo Methods” by Raymond Kim, supervised by Sanjoy Baruah
“Merging 360˚ Capture with 3D Reconstructed Environments for Improved Immersion in VR” by Vijay Rajkumar, supervised by Henry Fuchs
“Shared Haptics and Skeletal Reconstruction in a Social VR Training Simulator” by Nicholas Rewkowski, supervised by Henry Fuchs
5:00 – Session 3
“Rendering an In-Browser, 360° Environment from Disjoint Live Camera Feeds” by Luke Fernandez, supervised by Diane Pozefsky
“An IMU-Enabled Camera Rig for Wide-Range Stereo Videography in the Field” by Amanda Lohmann, supervised by Tyson Hedrick (Biology) and Diane Pozefsky
“Early Postnatal Anesthesia Exposure Reduces White Matter Micro-organization” by Jeffrey Young, supervised by Martin Styner
5:45 – Session 4
“Exploring the Efficiency of First-Order Proving Methods” by Mark Molinaro, supervised by David Plaisted
“An Architecture for Supporting Opportunistic Collaboration” by Dayton Ellwanger, supervised by Prasun Dewan
6:30 – Reception
Presentation Abstracts and Presenter Bios
Synthetic Data for Crowd and Human Understanding
Anson Wong, supervised by Dinesh Manocha
Crowd data is important for computers to improve crowd understanding but it’s hard to obtain. The work presented aims to generate synthetic crowd data with a procedural framework and to enable computers to learn it.
Anson Wong is a senior international student majoring in computer science and interested in robotics and machine learning. Anson is currently working under the supervision of Dinesh Manocha.
Evaluation of the Performance and Cost of Cloud-Based Robot Motion Planning
Jonathan Lynn, supervised by Ron Alterovitz
Prior work has shown that the computation of a robot’s motion plan can be split between the robot’s own low-powered processor and a cloud-based high-powered compute-optimized server. We evaluate the empirical performance of this method and the new cost model accompanying it with the compute service offered by Amazon Elastic Compute Cloud (Amazon EC2) for a robot using 8 degrees of freedom to complete a task in a dynamic environment, comparing across different data centers and server instances with an investigation of the roadmap complexity, the network time, and the total time required to complete the task.
Jonathan Lynn is a senior Computer Science and Global Studies double-major from Chapel Hill, NC and a member of the Computational Robotics Group. He hopes to further his interests in robotics and machine learning by pursuing a PhD in Computer Science after taking a gap year upon graduation.
Graph-Based Patrol Techniques for Use in Video Games
Ryan Gibson, supervised by Diane Pozefsky
Currently, most video games with stealth mechanics have enemy characters follow simple “patrol-chase-search” behavior trees with fixed, predefined routes and/or patrol points. We propose and evaluate several alternative patrol techniques that utilize a underlying graph structure in order to promote more dynamic and engaging gameplay.
Ryan Gibson is a sophomore pursuing majors in computer science and mathematics. He is currently enrolled in Diane Pozefsky’s Serious Games course, and his research interests revolve around artificial intelligence and graph theory.
Variance Reduction in Global Illumination with Monte Carlo Methods
Raymond Kim, supervised by Sanjoy Baruah
Monte Carlo methods are a family of numerical methods that solves problems by producing an approximation through a statistical approach using both deterministic and stochastic processes. The naïve way to reduce variance in our approximation is to increase the number of samples we take; however, as the complexity of the problem grows, this method becomes inefficient. We look into a few common variance reduction techniques and how we can apply them in the global illumination problem.
Raymond Kim is a senior majoring in computer science and mathematics who is interested in probability and geometry. He will be returning in the fall as part of the B.S./M.S. program.
Merging 360˚ Capture with 3D Reconstructed Environments for Improved Immersion in VR
Vijay Rajkumar, supervised by Henry Fuchs
I have designed a system combining 360˚ capture with 3D reconstruction to improve immersion in real world VR content — taking advantage of the best of both forms to enable a user with both full mobility and high quality imagery. We discuss the goals and design of the system, my implementation, and its limitations.
Vijay Rajkumar is a senior majoring in computer science. Vijay is interested in applying technology to evolving forms of media and storytelling.
Shared Haptics and Skeletal Reconstruction in a Social VR Training Simulator
Nicholas Rewkowski, supervised by Henry Fuchs
We explore a form of multiplayer training simulation that consists of the calibration of multiple virtual areas such that multiple players can share the same physical and virtual space, as well as the objects in it. We also present the pose estimation of accurate skeletal bodies, such as those given by the Leap Motion, such that natural input can be used to interact with the scene.
Nick Rewkowski is a sophomore computer science major working with Ming C. Lin and Henry Fuchs in the GAMMA and Telepresence groups, respectively. His research interests include multi-modal immersive virtual environments, 3D graphics optimization, and haptic feedback for practical applications.
Rendering an In-Browser, 360° Environment from Disjoint Live Camera Feeds
Luke Fernandez, supervised by Diane Pozefsky
This thesis details a novel approach to the rendering of a real-time 360° in-browser environment from a set of disjoint live camera feed. Developed to give police officers better visibility and reporting in the field, this study and the lessons learned during implementation demonstrate progress toward solving a high-stakes limited resolution and affordability problem that occurs when rendering panoramas from today’s off-the-shelf 360° camera technologies.
Luke Fernandez is a technology specialist and business implementation and design strategist who currently works as a technical business analyst intern at the SAS Institute. Luke prides himself in developing beautiful solutions to complex problems, and seeks to use technology to make the world a better place.
An IMU-Enabled Camera Rig for Wide-Range Stereo Videography in the Field
Amanda Lohmann, supervised by Tyson Hedrick and Diane Pozefsky
We developed a rotatable camera rig for biomechanics research. Our set-up allows us to record footage for stereo videography while rotating the rig, and then remove the effect of that rotation on our visual data during analysis.
Amanda Lohmann is a senior majoring in computer science and biology. Amanda is interested in using cool technology to study animal behavior and environmental interactions, and next year she is going to graduate school in ecology at Duke.
Early Postnatal Anesthesia Exposure Reduces White Matter Micro-organization
Jeffrey Young, supervised by Martin Styner
In this study, we investigated the effects of multiple, short exposures with commonly used anesthetics across two distinct cohorts of rhesus macaques. Specifically, we were interested in assessing the impact on the white matter microstructure as a result of these repeated exposures using diffusion tensor imaging (DTI). These results provide insight into the connectivity changes that the brain undergoes and shed light on the basis for the cognitive deficits observed in prior studies.
Jeffrey Young is a senior quantitative biology (B.S.) and computer science (B.S.) double-major graduating with highest honors. Jeffrey currently does research under the direction of Martin Styner, investigating the effects of anesthesia exposure on early brain development.
Exploring the Efficiency of First-Order Proving Methods
Mark Molinaro, supervised by David Plaisted
Many automated theorem proving applications rely on the DPLL algorithm for deciding the satisfiability of a set of propositional logic formulae. For first-order logic formulae, ground clauses within the Herbrand universe may be exhaustively enumerated below an incrementing size-bound and fed as input to DPLL. From even a cursory investigation of these enumerated clauses, it is evident that many of them have multiple repeated terms. Here, we explore a potential method for exploiting the size-bound by “cheating in” larger clauses with many repeating terms that may be relevant to the proof.
Mark Mollinaro is a junior computer science major and research assistant within the Carolina Institute for Developmental Disabilities. He is left-handed and his favorite candy is M&M’s.
An Architecture for Supporting Opportunistic Collaboration
Dayton Ellwanger, supervised by Prasun Dewan
This presentation gives an architecture to support opportunistic collaboration and an implementation of this architecture to aid in the instruction of computer programming courses. The work flow the architecture supports is very general and many versions exist outside of the context of collaboration in computer programming; the architecture and implementation were designed to be reusable in these other scenarios as well as the driving problem.
Dayton Ellwanger is a senior majoring in computer science and physics and minoring in mathematics. He has been working on this project for the past two years and is excited to finally see it in a mature state.