textfiles/computers/CYBERSPACE/delourau.ncr

Published in Dec 1992 CyberEdge Journal
	CyberEdge Journal is published by Ben Delaney, bdel@well.sf.ca.us


A SUMMARY OF VIRTUAL ENVIRONMENTS RESEARCH AT UNC-CHAPEL HILL

by Mark A. DeLoura, deloura@cs.unc.edu

The University of North Carolina at Chapel Hill's Computer Science
department has been doing research into immersive head-mounted virtual
environment systems since 1986, when their first head-mounted display
prototype was completed.  Since that time, one of the major goals of the
department has been improving the realism of virtual worlds, by advancing
the state of the art in both software and hardware systems.

In this article I'll briefly outline UNC's concentrations for the year, as
well as describe the current system used for developing VR-based
applications.

Fall 1992 sees the continuation of work on building PixelFlow, the newest
machine in a line of graphics multicomputers built by members of the
department.  PixelFlow, detailed in a SIGGRAPH '92 paper by Fuchs and
Molnar, will combine partial images produced by multiple independent
rendering pipelines in a high-speed image composition network to produce
the final image.  Performance of this machine is expected to be linearly
scalable to well over 10 million anti-aliased, textured polygons per
second, supporting advanced shading models and multiple shaped light
sources.  A working prototype of the PixelFlow system is expected to be
operational by early 1994.

The current rendering machine used by most VR-based applications in the
department is Pixel-Planes 5.  The Pixel-Planes 5 multicomputer was part of
the equipment brought to SIGGRAPH '91 by UNC, and was the graphics
workhorse used in all of the demos that were shown there.  (For more
information on the SIGGRAPH '91 "Tomorrow's Realities" demos, see CyberEdge
Journal issue #5.)  Pixel-Planes 5 is programmed in C or C++ with a subset
of PHIGS+, and can produce in excess of 2 million Phong-shaded, z-buffered
triangles per second.  VR applications are most commonly built using
various libraries created by students, such as PPHIGS (graphics), trackerlib
(tracking mechanisms), adlib (analog/digital devices), and vlib (virtual
world-specific routines, such as maintenance of standard transformations).

The Tracking group has developed a working optoelectronic tracking ceiling,
made up of many 2- by 2-foot ceiling tiles with 32 infrared LEDs per tile.
The special head-mounted display used with this ceiling tracker has four
cameras attached to it which point at the ceiling-- these provide enough
information for the computer to resolve the user's position to within 2 mm,
and orientation to 0.2 degrees.  Update rates depend on the mode the
ceiling is in, but 50-80 Hz is typical, as is a lag of 15-30 ms.  The
ceiling is currently 10- by 12-feet, but plans are in the works to increase
the size of the ceiling to 15- by 30-feet.  Research is underway to
develop a Self-Tracker, which can determine changes in position and
orientation by viewing the existing environment.

Head-mounted displays (HMDs) used by the department include a see-through
prototype, a video-merge HMD, VPL EyePhones, and the Virtual Research
Flight Helmet.  For more complex user interactions, a variety of
manipulators are available for use; these include an Argonne Remote
Manipulator (ARM) force-feedback arm, a billiard ball, a Python joystick, a
modified bicycle glove, a "wand", and a pair of analog joysticks.  All of
the hand-held input devices and HMDs (except for the optoelectronic tracking
ceiling) are tracked by Polhemus 6-D magnetic trackers (3SPACE and FASTRAK
models).

Work on software for improving the stability of virtual environments this
year is being led by Gary Bishop and the HMD group.  This year's motto is
"No Swimming", where swimming refers to the manner in which objects in
virtual worlds appear to slosh around when the user turns their head.
Swimming is the visible result of tracker lag, latency in the rendering
pipeline, and other bottlenecks in the system.  Several different areas are
being actively worked on to improve the images we see in the head-mounted
display: motion prediction using Kalman filters, beam-racing and
"just-in-time-pixel display" to get rid of the inaccuracies due to the
image-scanout time, examination of static and dynamic jitter in the
trackers, and correction of the distortion in the HMD due to the optics
used to achieve a wide field-of-view.

Aside from the war on swimming objects in virtual worlds, there are several
applications actively being worked on. The three major application projects
at this time are the Nanomanipulator, the ultrasound volume-visualization
project, and the architectural walkthrough. 

The Nanomanipulator, Russell Taylor's projected Ph.D dissertation topic, is
a joint project between the UNC Computer Science Department and the UCLA
Chemistry Department.  UCLA provided a Scanning-Tunneling Microscope (STM),
which Russell has created an inclusive interface to so that one can don the
HMD and actually change the surface of an object on a molecular level, as
well as feel the forces of the molecules via the ARM.  The display will
come up on an HMD or projection screen with cross-polarized shutter
glasses, and the user can interact with either the ARM or the billiard
ball.  The hand-input device has various modes attached to it, which
include feeling the surface, zooming in on a certain part of it, or
altering it.

The ultrasound project was shown in a paper at SIGGRAPH '92.  The
department has acquired an old ultrasound machine, and the goal is to be
able to construct a volume-visualization of the object being examined,
which would then be overlayed on top of live video and viewed with an
HMD.  This would make it seem as if a person had X-ray vision.  Testing is
commonly performed on a baby doll lying in an aquarium in the center of the
graphics lab, but tests with live subjects have been performed as well.
Closely associated with this project is the difficulty of overlaying
computer-generated imagery on top of the real world.  The real world is
inherently real-time, while the computer-generated objects are going to be
a bit slower due to the various bottlenecks of the tracking and
image-generation systems.  Different approaches for this application are
being examined, such as using a see-through HMD instead of viewing the
image overlayed on live video.

The architectural walkthrough originally was not in the plan for work this
year, but this decision was changed when it was pointed out as the only
application being worked on by UNC that made it apparent when
graphics algorithms were incorrect.  Most people have never seen surfaces
at a nanometer scale, or complex protein molecules, whereas an indoor scene
is something which nearly everyone experiences for large durations each day.
This makes debugging the shading models developed for use on the new
graphics machines easier to debug, since almost anyone can look at an image
and tell whether or not it appears realistic.  This year's approach to the
walkthrough deals largely with modelling details.  Through a cooperation with
Virtus Corporation, the Walkthrough project team is developing a much more
intricate model of the upcoming expansion of Fred Brooks' house.  The
models are created on Virtus Walkthrough software for the Macintosh, and
they are then uploaded to a Unix machine and converted to a Pixel Planes
5-specific format.  It is the hope that this new model will also be a great
test for the upcoming PixelFlow machine.

Other work being pursued at this time includes the addition of Focal Point
software for producing directional sound, inclusion of TiNi ticklers for
tactile feedback, expansion of the current 3DM inclusive world-building
tool, and continued work on Richard Holloway's excellent vlib package.