Not Nearly Enough Free Time

Dissertation

My dissertation focuses on the nanoManipulator, an interface to an Atomic Force Microscope that uses virtual reality techniques to give scientists deeper understanding and better control of their experiments. Although the nanoManipulator has been in regular use for six years, producing results in physics, chemistry, and biology, it is not well-suited to being run over the Internet. As the scientists we work with produce good results with our tools, they have attracted off-site collaborators. We have undertaken to give these distant scientists network access to our equipment and experiments, creating a collaboratory. As a NIH National Center for Research Resources, we have funding to determine the usefulness of this "distributed" approach to science, buying our collaborators computers rather than plane tickets.

In my dissertation, I address the problem of distributing our application over today's Internet -- a wide-area network without Quality of Service guarantees. Although bandwidth and loss are concerns, the critical problem I am attacking is latency. There are three levels at which one can discuss latency. Ideally, I would like to reduce latency as much as possible. So, at the lowest level, I ask how the application and the network can adapt to one another's state to ameliorate congestion, which is one of the principal causes of latency. Is a scientific application like the nanoManipulator amenable to the same kind of multi-dimensional adaptation as audio and video streams? If we can't reduce latency, I want to restructure the application to be more tolerant of latency. What is the best distribution of functions between sites? What alternate display and control methods can mask latency? If all else fails, I ask how to make the user explicitly aware of latency. Can a human be trained to adapt to latency?

From the specific problem of the distributed nanoManipulator, my dissertation generalizes to broader problems in distributed systems. How can adaptive applications be structured? What are the tradeoffs involved in using particular adaptation strategies, and how can we tell whether they are worth implementing? What techniques have been considered for other problems that should be moved across the ``wall'' between disciplines into the common vocabulary of researchers in distributed visualization?

Beyond Audio and Video: Multimedia Networking Support for Distributed, Immersive Virtual Environments. K. Jeffay, T. Hudson. To appear in Proceedings of Euromicro 2001.
Experiments in Best-Effort Multimedia Networking for a Distributed Virtual Environment. T. Hudson, M. C. Weigle, K. Jeffay, R. Taylor. Proceedings of SPIE Multimedia Computing and Networking 2001.
Enabling Distributed Collaborative Science. T. Hudson, D. Sonnenwald, K. Maglaughlin, M. Whitton, R. Bergquist. Video Program of ACM Conference on Computer-Supported Cooperative Work 2000.
In Situ Resistance Measurements of Strained Carbon Nanotubes. S. Paulson, M. Falvo, N. Snider, A. Helser, T. Hudson, A. Seeger, R. Taylor, R. Superfine, S. Washburn. Applied Physics Letters, Vol. 75 No. 18, 1 November 1999.
The Virtual-Reality Peripheral Network (VRPN) System. R. Taylor II, T. Hudson, H. Weber, J. Juliano, A. Seeger. UNC Chapel Hill Tech Report TR01-20.
Concurrency Control for Collaborative 3D Graphics Applications. T. Hudson. UNC Chapel Hill Tech Report TR01-021.

Teaching

Minicourses
I taught two minicourses on VRPN in Fall 99 and have reoffered them on demand in subsequent semesters. Some of the PowerPoint slides have been converted to HTML.
COMP 96: Computers and Society (Summer Session I, 1997)
Although taught in the Computer Science department, this class counts as a philosophy for undergraduates seeking breadth. I taught this class using the curriculum written by J. Walsh, extended to reflect my readings and the interests of the students. The textbook was the packet of photocopies used to teach the class during the year, with "Computerization and Controversy: Value Conflicts and Social Choices" by Kling and "Civilizing Cyberspace" by Miller as references. Although I had 35 students, I attempted to teach this class using a mixture of discussion and lecture techniques, as well as a few of the more unusual activities advocated by the Center for Teaching and Learning here at UNC-CH.

Other Research

Massive Model Rendering
For most of two years I worked with the Walkthrough group, half of that time as team leader. I did some development work on the system we showed at Siggraph 97, and spent the following Christmas Vacation taking a break from my disseration to help write the I3D paper.
- MMR: An Interactive Massive Model Rendering System Using Geometric and Image-Based Acceleration. D. Aliaga, J. Cohen, A. Wilson, H. Zhang, C. Erikson, K. Hoff, T. Hudson, W. Stuerzlinger, E. Baker, R. Bastos, M. Whitton, F. Brooks, D. Manocha. Proceedings of ACM Symposium on Interactive 3D Graphics 1999.
Occlusion Culling
Believing that there had to be an algorithmically simple approach to conservative polygonal occlusion culling (detecting polygons hidden from view by nearer opaque polygons) I developed a system in the summer of 1996; a paper on the results was presented at the Symposium on Computational Geometry 1997. The contemporaneous work of Teller and Coorg at MIT is a different approach with comparable results, and Hansong Zhang's method runs in hardware and in image-space instead of object-space.
- Visibility Culling using Hierarchical Occlusion Maps. H. Zhang, D. Manocha, T. Hudson, K. Hoff. Proceedings of ACM Siggraph 1997.
- Accelerated Occlusion Culling using Shadow Frusta. T. Hudson, D. Manocha, J. Cohen, M. Lin, K. Hoff, H. Zhang. Proceedings of ACM Symposium on Computational Geometry 1997.
Collision Detection
In late 1996, I was following the development of the VRML 2.0 specification perhaps more closely than anybody in the department; this led to my working with Ming Lin and the collision group here to integrate two previous approaches (RAPID and I-COLLIDE) and to propose an extension to the VRML standard to support collision detection between objects. The work was published at VRML'97. Inter-object collision detection remains a frequently-requested feature for VRML that I haven't yet seen generally implemented.

V-COLLIDE: Accelerated Collision Detection for VRML. T. Hudson, M. Lin, J. Cohen, S. Gottschalk, D. Manocha. Proceedings of ACM Symposium on Virtual Reality Modeling Language 1997.

Academic Projects

Fast Rendering of Caustics
For Turner Whitted's class, I looked at the Watt & Watt approach to caustics and said, "I can do that in hardware." On a RealityEngine2 and R4400 CPU, without optimizations, using the accumulation buffers and stencil buffers you can do 10FPS for a 30k pixel window with a single moving light casting reflections and refractions through 800 facets.
Collaborative Virtual Reality
While seeking a thesis topic in collaborative virtual reality, I looked at the problem of intercontinental design. If latency is unavoidable in a connection (e.g. Belgium to USA, at Caterpillar), and design teams are working in the same space but are not in constant communication or collaboration (e.g. submarine designers at Newport News Shipbuilding), then locking objects before modifying them becomes an unacceptable overhead. In a project spurred by Prasun Dewan's class I extended the theory of transformational concurrency control (the OPT/adOPTed algorithms) to virtual environments.
- Transformational Concurrency Control for Collaborative Design and Arrangement. T. Hudson. UNC-CH CS TR99-025.
Data Tablet
In Gary Bishop and Vern Chi's experimental Hardware Systems class, I helped build a data tablet using a new fabrication technique that avoids most of the existing patents, is very low-cost, and gives extremely good position data. (None of the teams tried for orientation or pressure; it's been posited that we could get one or both if we used a different architecture.) Only one of the four teams got good results, but we learned a few things. No details here until the originators are sure they aren't going to patent.

Trivia

Control Theory
While in BME 132 I wrote lots of matlab routines for my control theory work; some of them were posted here for other class members.

Last updated 23 February 2001.