Motivation for HCI Research

"Techniques and tools for interacting with virtual environments are at the core of research and development efforts around the world." Thus begins the Introduction to a Special Issue of the ACM Transactions on Computer-Human Interaction on Virtual Reality Software and Technology [September 1995].

During the past several years, virtual environments have gained broad attention throughout the computing community. During roughly that same time period, usability of the user interface has become a major focus of interactive system development.

Despite intense and wide-spread research in both VEs and in user interface usability, the exciting new technololgy of VEs has not yet been closely coupled with the important characteristic of usability--a necessary coupling if VEs are to reach their full potential. Appropriate development and usability evaluation of VEs can help prevent, for example, the situation in a VE where there is a fundamental mismatch between what a device is suited for versus what it is actually being used for.

Although numerous methods exist for developing interactive computer applications, these methods have well-known limitations, especially when applied to VEs. For example, many existing usability evaluation methods are time-consuming and personnel intensive. Most methods are applicable only to a narrow range of interface types (e.g., GUIs--graphical user interfaces) and have had little or no use with innovative, non-routine interfaces such as those found in VEs. VE applications have interaction styles so radically different from ordinary user interfaces that traditional user-task-based development may be neither appropriate nor effective. The focus of most existing methods, while properly user-task-based, is on a single user performing isolated, low-level user tasks--very different than VEs in which one or more users are performing integrated, shared, multi-thread tasks. For example, VE tasks such as navigation, distance estimation, depth perception, object manipulation, and visual search need interaction techniques unlike those typically used in GUIs. Many characteristics that generally are unique to VE applications and are key to their usability, such as perceived presence and fidelity, are absolutely not addressed in existing development methods. Methods do not support, for example, quantification (or even qualification) of a user's perception of such characteristics. Methods do not easily support evaluation of VE applications in which two or more users interact; this requires evaluation, as a whole, of multiple users using different sets of hardware, perhaps even at different physical sites. Immersive versus non-immersive views are another characteristic of VE interaction that does not exist in traditional user interfaces.

Too little attention to usability of VEs is reported in the literature. More likely, unsubstantiated claims of improved performance and user satisfaction are based at best on a few user interface guidelines and at worst on warm fuzzy feelings of VE developers and "way cool" comments from VE users. The few reported evaluations are rarely comprehensive. For example, novel interaction techniques developed by van Dam et al. at Brown University are routinely evaluated for usability, but seldom are the techniques set in a realistic application and used to perform 'real' user tasks within that application. Thus far, VEs have been largely about "gee whiz" technology. An in-depth literature review by a Virginia Tech PhD candidate [Snow, 1995] confirms a lack of foundational work for VE development and especially for evaluation.

The field of VEs is now maturing to the point for basic research to be not only fruitful, but actually critical to optimal use of this promising technology. Evaluation of human performance and satisfaction lies at the heart of assessing and improving user interfaces to VEs. An underlying assumption among both researchers and developers seems to be that VEs, because they are a new technology, are inherently good and usable. Progress is needed to move beyond this flawed assumption, to have a focus on usability become a routine activity in VE development. Thus, a key aspect of our VE research is that we want to go beyond simply creating new technology and assuming it, merely by its novelty, is inherently better for a user than existing technology. Instead, we will continually interact with users to assess the effects of our new technology on user task performance and satisfaction. As interactive systems move beyond WIMP (window, icon, menu, pointer) interfaces, an evolutionary cycle of conceptualization, implementation, and evaluation becomes essential to produce VEs that are effective and efficient, not merely new and different.

Background: HCI Development

Human-computer interaction (HCI) is what happens when a human user and a computer system cooperatively perform tasks. Human-computer interaction includes user interface hardware and software, user and system modeling, cognitive and behavioral science, human factors, empirical studies, methodology, techniques, and tools. The goal of much work in human-computer interaction is, in one way or another, providing users with a high level of usability.

As shown simplistically in Figure 1, a fundamental dichotomy exists in user interface development [Hix & Hartson, 1993]: user interaction development and user interface software development. The user interaction component is how a user interface works, its "look and feel" and behavior in response to what a user sees and hears and does while interacting with the computer. The user interface software component is the implemented code that instantiates the interaction component. Both are necessary. However, much research, especially in VR applications, focuses on interface software development and explores very little about interaction development.

Objective and Approach for HCI Reserach

Despite many research advances in interactive computer systems, usability barriers still exist that impede human productivity and have a profound impact on computer users in business, government, industry, education, and indeed, the whole nation. To break down these usability barriers we need vastly improved development methodologies, focused on ensuring usability of the user interface. Capitalizing on our expertise in HCI research (see Why Us? section) and the uniqueness of our user interface research facilities (again, see Why Us?), we propose to use the CAVE(tm) to study use of immersive VEs. Our goal is two-fold. First, we wish to determine the most effective user interface models and metaphors to use in developing usable CAVE(tm)-based applications. Second, at a meta-level, we wish to improve the overall VE development process.

The methodological gains of human-computer interaction must be brought to bear as part of a new technology for breaking barriers to interactive computing. We propose to attack barriers on a broad scale, producing methods to change the way developers of VEs think about the process of developing these systems. We anticipate innovative, ground-breaking research leading to new ways of focusing on user interaction as a central characteristic of VE development and use.

One of the most important advances in HCI in recent years is a recognition that the process of user interaction development must be user-centered, focusing on the human rather than the system [Hix & Hartson, 1993]. Because of the dichotomy explained above, traditional, accepted methods for software engineering [e.g., Boehm, 1988] do not necessarily produce a user interface that is usable, useful, and effective [Gould & Lewis, 1985], [Whiteside et al., 1988], [Hartson & Hix, 1989], [Gould, Boies, & Lewis, 1991]. A development methodology that revolves around early and continual user involvement in the development process is the accepted approach to achieve a high quality user interface [Whiteside et al., 1988], [Hartson & Hix, 1989], [Gould, Boies, & Lewis, 1991]. This approach, which includes user analysis, user-centered requirements definition, rapid prototyping, formative usability evaluation, and iterative refinement [Hartson & Hix, 1989], is the one that will be employed and extended in this project to invent, develop, and study user interfaces for CAVE(tm) applications.

In particular, we will involve users from inception to installation of our targeted applications. From the beginning, we will interview a variety of potential users of CAVE(tm) applications such as the ones previously described, to determine what kinds of "objects" they work with, and--equally importantly for the user interaction design--how they would like to work with these objects. For much of our work, we will employ usability engineering techniques [Whiteside et al., 1988] to quantify usability of the CAVE(tm) applications. In particular, we will set usability specifications (operationally defined criteria for measuring usability) for our systems, and test representative users to find usability problems. We will collect both quantitative and qualitative user performance data, compare our results to the usability specifications, and perform cost/benefit analyses to determine which problems most affect system usability. These problems can then be resolved in order of their affects on usability, rather than in an ad hoc, random fashion. Using this approach will achieve both our goals: to determine which methods of interaction with CAVE(tm)-based applications are most effective for a user, and to improve the VE application development process.

Why Us?

Our Qualifications
The Human-Computer Interaction Project at Virginia Tech is one of the pioneering groups in HCI, in existence since 1979. Working in an interdisciplinary setting of computer scientists, human factors experts, psychologists, and psychometricians, our group has become internationally recognized for state-of-the-art contributions in conceptual areas including user interface development methodology, modeling of human-computer interaction, and software support tools for the interface development process. We have extensive research interaction with organizations such as IBM, DEC, the Naval Research Laboratory, Eastman Kodak, Naval Surface Warfare Center, PRC, NCR, and many others, giving us insight into real-world problems. This breadth of interdisciplinary research and application is rarely found in an academic setting.

Unique Physical Facilities
Our HCI group has had more than $4 million in research funding, including an NSF Research Infrastructure award ($2 million, with funds from Virginia Tech), specifically for building a unique HCI research infrastructure involving three laboratories:

• Usability Methods Research Laboratory, a unique lab designed specifically to develop, evaluate, and apply high impact methods for conducting formative and summative usability evaluations. We believe this is the only laboratory in the world specifically designed and instrumented for user interface development methods research, including VE application development.

• Interaction Technology Laboratory, a high-tech facility to house a broad variety of state-of-the-art hardware and software related to HCI, including VE, for development and demonstration purposes.

• Information Access Laboratory, an applications lab for studying multimedia information access with real-life, realistic-size document collections.

TARGET APPLICATION AREAS

Following are descriptions of three target application areas, as examples of the kinds of application-based research we will pursue in the CAVE(tm). The HCI research, just described, will be used to study these applications, as appropriate.

CAVE(tm) Visualization of Biomolecular Structures (Dr. David Bevan)

We are engaged in several projects in which computational techniques are applied to studies of protein structure and function. In one of these projects, we are computing free energies of isomerization of peptide bonds, the results of which will contribute to a better understanding of protein folding [Kurusu & Bevan, 1994]. In another project, we are simulating the dynamic properties of peptides, using published data from fluorescence resonance energy transfer to validate the computations [Bevan, 1995]. This analysis represents a novel application of molecular dynamics to biomolecules, and from it we will be able to infer the atomic and molecular motions of peptides and understand better the origins of their motion and stability. In a third project, we are using homology modeling to provide initial models of two recently discovered proteins. One of the proteins is a protein phosphatase from an archaebacterium and thus may represent an ancestral form of homologous enzymes in higher organisms [Leng et al. 1995]. The other protein, called water-stress protein, has been isolated from another primitive organism (a cyanobacterium) and contributes to the desiccation tolerance of that organism [Scherer & Potts, 1989; Sines, 1996]. We also are conducting research to develop quantitative structure-activity relationships (QSAR) [Shannon et al., 1991; Bramble et al., 1994]. In this project, we are attempting to apply computational techniques to calculation of molecular parameters that may be effective descriptors in the QSAR. One of these is the free energy of solvation, a parameter that will affect membrane permeability of molecules (e.g., drugs) and the binding of ligands to their receptors (i.e., docking).

These projects currently are performed using software such as Quanta/CHARMm, MidasPlus, DOCK, AMBER,Gaussian94, MOPAC, and AMSOL. The computationally intensive programs (e.g., AMBER, CHARMm, Gaussian94) are installed on the University IBM SP-2. The others are installed on departmental workstations.

Central to these projects is the need to visualize the structures throughout the simulations. Currently, we view structural information on graphics workstations, while recognizing the limitations inherent in displaying a three-dimensional molecule in two dimensions. A CAVE(tm) will enhance our ability to visualize the molecules, allowing us to take a significant step forward in our research efforts. A CAVE(tm) also will provide the environment to gain a better "feel" for molecular interactions through tactile feedback, thereby enhancing our molecular modeling capabilities. These vastly improved resources will revitalize our research by allowing us to undertake projects currently impossible.

We will begin our CAVE(tm) experience by collaborating with researchers at NCSA. By traveling to NCSA, Bevan and students will quickly gain the benefit of CAVE(tm) visualization and then return with expertise that will enable more rapid implementation of CAVE(tm) technology at Virginia Tech.

We also will integrate the CAVE(tm) into our curriculum. For example, Bevan teaches a graduate-level course entitled "Molecular Modeling of Proteins and Nucleic Acids." Students learn the theory of molecular modeling, but much of the course is taught from a more practical perspective in that students are required to do several computer-based problems and projects. The students in the molecular modeling course will all experience the CAVE(tm) for visualization, and those with the inclination may apply it in their class projects. The Department of Biochemistry at Virginia Tech has the second or third largest undergraduate enrollment (~300 students) in the country. Bevan will develop programs to use the CAVE(tm) to assist these students in visualizing protein and nucleic acid structures as part of their undergraduate training. Many years ago when we first began using computer visualization of structures in our classes, the students were very enthusiastic because they gained a much better understanding of structural features. The CAVE(tm) will raise this educational experience to a level that far surpasses what we are currently able to provide.

Failure and Reliability in Fiber-Reinforced Metal and Ceramic Composites (Dr. W. A. Curtin)

There are three requirements for the successful use of ceramic and metal matrix composites in high-temperature structural applications in advanced jet engines and energy conversion systems. First, the microstructure of the material must be optimized to perform as designed; second, the intrinsic reliability of the as-designed materialmust be known so appropriate safety factors can be set; third, the as-designed material must be manufacutred.

All three key requirements thus depend on the micromechanics of broken fibers in the composite and on the effects of stress concentrations on the progression of fiber breaks in the composite. Our research is specifically aimed at quantitatively addressing these issues using numerical computer simulation models of damage evolution, including the realistic 3D damage aspects.

We have developed a new numerical technique to investigate the effect of stress concentrations on performance as a function of the underlying micromechanical properties of the composite constituents [Zhou & Curtain, 1995]. The computational technique uses a highly efficient algorithm based on Lattice Green Functions to calculate the 3D stress distributions in aligned fiber bundles for arbitrary configurations of broken fibers in the composite, and determines the evolution of specific fiber damage configurations throughout the material as increasing loads are applied. The numerical simulations produce composite strength distributions in the absence of manufacturing defects and, using visualization techniques, can demonstrate the specific sequence of 3D fiber damage that leads to development of a complex "critical defect" that causes composite failure.

While the numerical simulation technique and the micromechanics are now well established and a wide variety of calculations are being performed, a major limitation in the complete analysis of the composite failure process is proper and full 3D visualization of the damage state of the material. Although the numerical calculations are in full 3D, visualization of the evolving stress states is currently restricted to sequential observation of 2D slices of the composite perpendicular to the fiber/tensile load axis. The analysis of 2D slices, and the reconstruction of appropriate "critical" defects in 3D, is very time consuming and not amenable to a clear physical analysis of the failure evolution. The failure evolution will occasionally take place on a different plane from an identified plane that looks as if it will become critical.

In addition, current identification of the critical region involves running a complete simulation to locate the central failure plane and then a repeat simulation to retain damage evolution in the 3D vicinity of that plane, thus demanding twice the computational time. The overall data storage requirements of the full progression of damage throughout the entire material and over the full loading history is also prohibitive at the present time. Under certain conditions relevant to many practical composites, the critical defect that drives failure is a diffuse 3D distribution of break clusters which are mechanically unstable and drives macroscopic component failure. The 2D visualization of such clusters and their evolution is quite unsatisfactory. Furthermore, the analytic models we are developing based on the simulation results invoke the concept of a generalized "critical cluster" and have specific implications for the size of such clusters; accurate determination of the critical clusters from 3D visualization of the damage would greatly aid in assessing the applicability of the analytic concepts to the real materials.

We can easily introduce random damage into the simulation models, simulate the failure, and then reconstruct the material and investigate the vicinity of the critical damage initiation. However, identification of the initial damage and damage evolution are really rather difficult to study from 2D images only. Other types of manufacturing damage are also being studied currently, and visualization of the damage evolution, and in particular the localization of damage around large initial defects, would be particularly useful as this work progresses.

We have attempted to develop 3D representations of the stress/damage state in the composite, but the current 3D representations on a 2D terminal screen are far too complex to study in detail. Moreover, the spatial relationships of various fibers are difficult to assess because of shielding and the sheer number of individual elements requiring attention. Our current composite sizes are 400-1000 fibers in the cross-section with 20-40 elements per fiber in the longitudinal direction. And such sizes are the minimum sizes required to obtain proper results. The stress states in such a system vary rapidly from fiber to fiber, and vary fairly rapidly over the length elements as well. Since fiber failure is driven by specific high stresses on just one or a few elements, complete resolution of the entire 8000-40000 elements is needed to accurately capture the damage state.

In light of the clear advantages that visualization provides for observation and interpretation of composite damage evolution and failure, a highly advanced, 3D visualization environment will be extremely helpful in furthering our research in this area. The CAVE(tm) provides, in particular, the possibility of entering into our 3D composite and searching out complex damage states internally. An ultimate real-time and/or interactive mode of operation would allow damage evolution to be fully followed during the computation, and would eliminate the need for multiple CPU-consuming identical runs and for tedious and difficult reconstruction of composite damage states from 2D slice data. As 3D Computer-Aided Tomography becomes even more practically available, the simultaneous visualization of damage in actual components and in model composite representations can also be coupled through the use of visualization technology.

In summary, the complexity of the failure processes in such advanced, heterogeneous, materials requires 3D visualization to fully interpret and utilize the information contained in the simulation models and to couple that information to real material systems. Acquisition of the CAVE(tm) will thus ultimately assist us in guiding manufacturing process, NDE interpretation, and providing reliable predictions of the performance.

CAVE(tm) Visualization of Results of Large Scale Simulations in Material Science (Dr. Diana Farkas)

This project involves a basic study of the structure of various defects in materials at a molecular atomistic level. The project uses descriptions of interactions between the various atoms that are based on both experimental data and the first principle quantum mechanical calculations. The technique we presently use for interatomic forces is the embedded atom method (EAM) [Daw & Baskes, 1984] and a modified version of the embedded defect (ED) method [Pasianot et al., 1991] that we have developed.

Currently we are studying various intermetallic materials that are novel candidates for structural applications at high temperatures. We have developed force laws (interatomic potentials) for a series of materials [Farkas, 1994a; Farkas et al., 1994] and are now working on the description of impurity effects in these materials. Once the force law description for all component elements is obtained, the equilibrium atomic configuration around any defect can be calculated through various energy minimization schemes, such as molecular statics, molecular dynamics, or Montecarlo techniques. Detailed information on atomistic configuration of the defective region of the solid is then linked to properties in the material that are controlled by such defects. The main defects that we are interested in are interfaces, dislocations, and cracks. These defects control the mechanical behavior of structural materials.

The Atomistic Simulation Laboratory at Virginia Tech has been involved in atomistic computer simulation of materials behavior for more than 10 years. Our work has included contributions in the area of grain boundary structure [Farkas, 1994b], dislocation core structure [Ternes et al., 1995], planar faults [Farkas & Vailhe, 1993], point defects [Xie & Farkas, 1994], and cracks [Shastry & Farkas, 1995]. In recent years the increased speed of available computing facilities has allowed us to undertake large scale massive simulations involving millions of atoms and study to structures of defective solids and their relation with material properties.

The Atomistic Simulation Laboratory has facilities consisting of various workstations networked with supercomputing facilities in the College of Engineering (SGI Power Challenge) and the University ( IBM SP-2). For visualization we currently use the Scientific Visualization Laboratory. For example we have developed techniques to visualize dislocation core structures. These large scale simulations pose new challenges in ways in which results can be visualized and analyzed. Conventional two-dimensional scientific visualization packages usually cannot handle the large number of atoms involved in these massive simulations. Furthermore, new computer architectures and increasing speeds allow the study of defects that are truly three-dimensional in nature. This means that the results have to be visualized necessarily in three dimensions. An example of this type of simulation is work on the study of cracks in intermetallic materials currently under way in the Laboratory. This work is now possible in three dimensions using realistic crystal structures for the materials considered, and preliminary results indicate that the benefits of CAVE(tm) visualization on the fracture mechanics studies at an atomistic level are indeed going to lead to qualitatively new possibilities. Similar benefits are expected in studies of various other defects underway in our laboratory, such as the structure of stepped surfaces, grain boundaries, and interfaces in various materials. Our current work is sponsored by NSF and ONR, representing a large effort with a group of two postdoctoral research fellows and six graduate students. In our proposal we will develop techniques for visualizing crack propagation in three dimensions. Simulations will be carried out for a series of different intermetallic compounds of technological relevance, including single phase materials, ( i.e., NiAl) and two phase materials ( i.e., two phase intermetallic materials in the Ti-Al system)

We will start with CAVE(tm) applications already in place at NCSA and we will start training graduate students in use of the CAVE(tm) as part of the current projects. Preliminary discussions on a joint project with NCSA have been initiated for this purpose. We will start by visualizing results from static simulations and will later on develop dynamic studies and visualize the dynamics of crack growth. With our CAVE(tm) we will progress to doing real time visualization of dynamic fracture behavior. The main interest will be to study the ductile or brittle response of these materials. This is the essential point from a basic science view, since these studies will contribute to the fundamental understanding of fracture processes. This is also an essential point from a technological point of view, since brittleness is the main limitation for realistic descriptions of energetic interaction among atoms. Molecular statics and molecular dynamics studies of crack behavior require complex visualization schemes to translate the three-dimensional structure into various two-dimensional plotting schemes. CAVE(tm) visualization will allow this research to take a qualitative step forward since the structure of the crack and surrounding area can be seen in a direct way.

During a recent visit to NCSA's CAVE(tm), a sample simulation was already visualized in the CAVE(tm), using applications already in place. These preliminary results indicate that the benefits of CAVE(tm) visualization on the fracture mechanics studies at an atomistic level are indeed going to lead to new possibilities. In this work we improve on the visualization techniques that we have developed at the Visualization Lab for the various defects studied over the last ten years.

We propose to integrate results of this research into both our graduate and undergraduate curriculum. The impact of CAVE(tm) molecular modeling visualizations on our graduate curriculum in materials science will be through courses on Computer Simulation in Materials, Science, Dislocation Theory, Diffusion Processes and Fracture of Materials. All these graduate classes are currently being taught by Professor Farkas. In our undergraduate curriculum the simulations will be incorporated in our sophomore materials science class and in our sophomore class on Analytical Methods in Materials Science, taught by Professor Kriz.

D. R. Bevan. Virginia J. Sci. 46, 106 (1995).
B. Boehm. A Spiral Model of Software Development and Enhancement. IEEE Computer. (1988).
L. Bramble, G. Boardman, D. R. Bevan, & A. Dietrich. Environ. Toxicol. Chem. 13, 307 (1994).
J.M. Carroll, C.A. Shaffer, M.B. Rosson, J.K. Burton, L. Arrington, NSF-NIE Grant, 1996-1999. (1995).
M. Daw & M. Baskes, Physical Review. 29, 6443 (1984).
M. Duva, W. A. Curtin, & H. N. G. Wadley. Acta Metallurgica 43, 1119 (1995).
D. Farkas. Metallurgical Transactions. 25A (1994).
D. Farkas. Modelling and Simulation in Materials Science and Engineering 2, 975 (1994).
D. Farkas. B. Mutasa, C. Vailhe, & K. Ternes. Modelling and Simulation in Materials Science and Engineering 3, 201 (1994).
D. Farkas & C. Vaihe, J. ofMaterials Research, 8, 3050 (193).
J. Gould, S. Boies, & C. Lewis. Making Usable, Useful, Productivity-Enhancing Computer Applications, CACM 34 ( 1991).
J. Gould & C. Lewis. Designing for Usability--Key Principles and What Designers Think. CACM. 28 (1985).
H. R. Hartson & D. Hix. Toward Empirically Derived Methodologies and Tools for Human-Computer Interface Development. Int'l J. Man-Machine Studies 31 (1989).
D. S. Hix & H. R. Hartson. Developing User Interfaces: Ensuring Usability through Product & Process. John Wiley & Sons, Inc. New York. (1993).
M. Ibnabdeljalil & W. A. Curtin. Submitted to Intl. J. of Solids and Structures. T. Kurusu & D. R. Bevan. Virginia J. Sci. 45, 216 (1994).
J. Leigh, A. Johnson, C. Vasilakis, T. DeFanti. Multi-perspective Collaborative Design in Persistent Networked Virtual Environments, to appear Proceedings of VRAIS 96. (1996).
J. Leng, A. J. M. Cameron, S. Buckel, & P. J. Kennelly. J. Bacteriol. 177, 6510 (1995).
R. Pasianot, E. Savino, & D. Farkas. Physical Review. 43, 6952 (1991).
S. Scherer & M. Potts. J. Biol. Chem. 264, 12546 (1989).
R. D. Shannon, G. D. Boardman, D. R. Bevan, & A. M. Dietrich. Environ. Toxicol. Chem. 10, 57 (1991).
V. Shastry & D. Farkas, presented at the MRS Fall Meeting (1995).
B. Shneiderman. Designing the User Interface: Strategies for Effective Human-Computer Interaction. Addison-Wesley. (1987).
B. J. Sines. Isolation and Partial Characterization of a Water Stress Protein of the Desiccation-Tolerant Cyanobacterium Nostocmune UTEX 584 Expressed in Escherichia coli., M.S. Thesis, Virginia Tech. (1996).
K. Ternes. Z-Y. Xie & D. Farkas, Materials Science and Engineering, Vol. A192/193, 125 (1995).
J. Whiteside, J. Bennett, & K. Holtzblatt. Usability Engineering: Our Experience and Evolution. Ch. 36 in Handbook of Human-Computer Interaction, M. Helander (ed.). Elsevier. (1988).
Z. Y Xie & D. Farkas. Atomistic Structure and Lattice Effect of Vacancies in Ni-Al Intermetallics. J. Mater Res. 9, 875 (1994).
S. J .Zhou & W. A. Curtin. Acta Metallurgica. 43, 3093 (1995).

Together the CAVE(tm) hardware and the user interface software constitute the instrument. In this project the instrument is more than just a hardware item but rather the instrument is the user interface in the same sense as the network is the computer. The user interface becomes the medium that allows engineers and scientists to transform their data-rich information-poor world into an information-rich experience. Because the user interface is an inseparable component of the CAVE(tm) instrument, we emphasize the need for a CAVE(tm) HCI user interface programmer. Thus a user interface programmer is included in the budget much the same as a component of the instrument. Other cost shared support staff are discussed in more detail in the Project and Management Plans on page C.2.12.

Both the HCI user interface group and the simulation-visualization applications group want to avoid the time consuming task of constructing our own home grown virtual environment. Because both groups need a fully functional CAVE(tm) instrument to support existing research programs, we chose to purchase the CAVE(tm) from Pyramid Systems Inc. The CAVE(tm) system purchased from Pyramid Systems Inc. ($175,000) includes the license to use the Electronic Visualization Laboratory (EVL) CAVE(tm) library but this cost does not include the purchase of the SGI Power ONYX. It is important that although engineers typically enjoy building systems, we will not attempt to build a CAVE(tm) but rather this proposal focuses on equipment acquisition and training.

Details of the standard CAVE(tm) equipment configuration is given by Pyramid Systems Inc. and is similar to the CAVE(tm) system configuration at the National Center for Supercomputing Applications (NCSA). The largest part of the cost are the video projectors. The purchase of equipment from Pyramid Systems Inc. does not include the required SGI Power ONYX . To keep the cost to a minimum we choose to initially create a three-walled CAVE(tm). Hence a Power ONYX with three reality engines, is configured with 16MB texture memory, 2x90MHz R8000, 64MB memory, 1 IMB, 2GB system disk, first 512MB Super Dens Mem, 2 way interleave, IMB, two RealityEngine2 Graphics subsystem for GR systems, includes 1RM and 16MB texture memory, nine addition 16 MB Texture Memory Raster Manager (RM) card for RealityEngine2, Graphics Expansion Card Cage, upgrade 2GB to 4.3 GB SCSI-2, and finally three MultiChannel Option with Break Out Box, a 21-inch MultiSync Monitor, IRIX 6.1 Operating System Software and Manu- als, first years maintenance: ($780,303) this price includes an educational discount. Note that this a minimum system configuration where most CAVE(tm) systems also purchase a low end workstation such as an SGI Indy to remotely control the Power ONYX.

Existing simulation-visualization equipment on campus is now outdated. The fast moving market of supercomputing for performing faster calculations and visualizing results has recently become evident at Virginia Tech. Recently the University has purchased an IBM SP2 and the College of Engineering has purchased a SGI Power Challenge. Both computers have been purchased for compute cycles only, hence Engineering has named their machine "Crunch". Desktop workstations are not capable of visually processing simulation results in realtime. Although not all simulation results require realtime visualization in a CAVE(tm) format, most applications can benefit from visual postprocessing simulation results. A list of projects that are routinely using visual data analysis tools can be found on page C.2.1. For most of these applications it is important that the engineer / scientist can manipulate the CAVE(tm) environment with spatial positioning technology.

In the second and third years, after the CAVE is fully functioning, we propose to work with a Spatial Position Systems Inc. (SPSI) located at the Virginia Tech corporate Research Center. SPSI has been involved with creating new 3D realtime position technology in the automation and construction industry. With SPSI as one of our first industrial partners, we proposed to work with SPSI to create multiple sensor arrays that would allow truely interactive virtual environments. For example with multi head mount positioning devices together with stereo displays, several CAVE(tm) participants could interact with each other as well as the visual CAVE display.

SPSI is also interested in experimenting with onsite use of positioning sensor arrays to simulate and evaluate the design of future SPSI positioning devices. Results of this research could than be marketed and sold to other industrial partners such as building and construction companies that would benefit from on-site "walkthrus". Developing new positioning technologies could be shared with our NCSA-SGI Power Grid Alliance partners.

It is important to note that development of new positioning technologies not interfere with the use of the CAVE(tm) for routine research.

For the last six years engineers and scientists at Virginia Tech have experimented with the use of simulation and visualization (SV) tools in their individual research projects. Many of these research projects have used the Laboratory for Scientific Visual Analysis and associated classes to experiment with turnkey visual data analysis tools such as AVS, PV-Wave, Molecular Simulation Inc., and others. From this experience we learned that there is a higher regard for visual thinking and as a result several projects have continued to explore the use of SV. Some of the projects that have benefited the most from the use of SV are listed on page C.2.1.

Most of those researchers who have continued to explore the use of SV tools have reached a limit either with respect to the visual data analysis software or hardware capabilities presently on campus. For example, recently the University purchased an IBM SP2 and the College of Engineering purchased a SGI Power Challenge both exclusively targeted to meet growing demand for faster calculations. Hence the SGI Power Challenge was given the name "Crunch". In turn the larger data sets are generated and now there is also an emphasis on more efficient use of visual data analysis tools (scientific visualization) to "Look" at these larger data sets.

Looking at larger data sets requires us to rethink how we use visual data analysis tools in our research. At Virginia Tech we have a unique opportunity in that we have a strong HCI group that can help us take the next step in rethinking how engineers and scientists can use visual data analysis tools to "Look" at their data. With respect to hardware the CAVE(tm) literally adds another dimension to visual data analysis tools. Existing HCI funded research can also benefit from access to CAVE(tm) technology. Specific target applications could also benefit from fundamental HCI CAVE(tm) research.

The details as to why the CAVE environment is essential to advancing research in both HCI user interface development and evaluation, and simulation-visualization applications in engineering and the sciences is justified and explained in some detail in the previous sections of this proposal. Although there are many applications listed on page C.2.1, the three targeted applications in Visualization of Biomolecular Structures, Fracture in Composites, and Visualization of Large Scale Simulations in Material Science where choose because they were the best candidates that could benefit from visual data analysis in a CAVE(tm) format.

These applications can also benefit from our recent NCSA SGI Power Grid Alliance.

The Laboratory for Scientific Visual Analysis has been an NCSA Academic Affiliate since 1991. Recently Virginia Tech has been accepted as a NCSA-SGI Power Grid Alliance member where we have formally established a partnership between our two institutions. The proposal has been posted at http://www.sv.vt.edu/future/proposal.html. This proposal is consistent with Virginia Tech's commitment to enhancing its teaching and research mission through the use of supercomputing, human-computer user interfaces and visualization in various research programs, undergraduate and graduate curriculum, and through extension to Virginia as a land grant university.

We anticipate that this alliance will benefit both institutions: (1) Virginia Tech benefits by access to current state-of-the-art CAVE(tm) technology at NCSA and (2) NCSA benefits by increasing the engineering and scientific applications base and most importantly access to a well established HCI user interface team. The HCI group at Virginia Tech is particularly well known for their focus on HCI evaluation. Because the CAVE(tm) is still a relatively new tool, there is much to be learned about how to effectively use features unique to a CAVE(tm) from an HCI perspective.

Because of potential benefit to the NSF Science Technology Center on High Performance Polymeric Adhesives and Composites, this NSF Center will provide $25,000 over the three years period that can be used to facilitate an exchange of key CAVE(tm) personnel from both institutions. Starting in the summer of FY97 these funds will be used to train key Virginia Tech faculty and staff to learn how to work with CAVE(tm) technology under the NCSA-SGI Power Grid Alliance.

The NSF Science and Technology Center will also contribute three Summer Undergraduate Research Program (SURP) positions ($45,000/3 years) where SURP students will be chosen from under-represented groups to work with CAVE(tm) technology as part of their summer research project. For the last four years Professor Kriz has work with under-represented students in the SURP program. A recent example of a SURP project is posted on the WEB at http://www.sv.vt.edu/class/surp/Yilma/MolecDyn/Molec.html. From SURP'94 two other under-represented students completed projects in molecular modeling and viscoelasticity. SURP student exposure to CAVE(tm) technology will provide new opportunities for these students in their future carriers.

Professor Ron Kriz on Simulation-Visualization (SV) applications and Dr. Deborah Hix on Human-Computer Interactions (HCI) will direct and coordinate the SV and HCI activities respectively proposed in the previous sections of this proposal. Professor Kriz is an expert in Scientific Visual Data Analysis where in the last six years he has directed activities in the University Laboratory for Scientific Visual Analysis (http://www.sv.vt.edu) and taught a course on Scientific Visual Data Analysis and Multimedia for the last six years: a summary of student projects are located at http://www.sv.vt.edu/class/Student_Proj/Student_Proj.html. Professor Kriz was the director of Digital's pilot Visualization Reference Center and has worked with faculty across campus: for example Professors David Bevan and Kriz were successful in establishing a software site license with Molecular Simulations Inc. Dr. Deborah Hix is a well-known expert in human-computer interaction, having worked in this field for more than 15 years. During this time she has supervised numerous students, both graduate and undergraduate, administered a broad variety of research contracts and grants, and had extensive experience in many aspects of user interface design and development.

The project management plan is: 1. establish a working CAVE(tm) system at the end of the 5th month. This is possible because the physical construction of the CAVE(tm) will be subcontracted by Pyramid Systems Inc. (see attachment 1) , 2. with the NCSA CAVE(tm) expert begin a training program that will be designed to get members of the HCI team and Drs. Bevan, Curitn, and Farkas and their graduate students CAVE(tm) literate and the remainder of the first year will focus on the "Molecular Modeling Design Project" , 3. second and third years will be a collaborative programing effort to convert existing visualization software tools into a CAVE(tm) format and with a working CAVE(tm) environment begin the proposed development of HCI user interfaces. We have used $60,000 of Virginia Tech funds for Dr. Hix to direct current HCI research in the CAVE(tm).

With respect to training, the CAVE(tm) will be used by graduate students for their class projects in the Scientific Visual Data Analysis class. Typically these class projects have been incorporated as part of Masters and Ph.D. dissertations. Other existing HCI classes will also take advantage of access to the CAVE(tm) to conduct HCI research. Another exciting training program is the NSF-STC Summer Undergraduate Research Program (SURP) where for the last four years Professor Kriz has worked with under-represented students on projects related to material science. In a recent SURP project for example,a minority student created a web tutorial on "Understanding Molecular Dynamics" which was posted at http://www.sv.vt.edu/class/surp/Yilma/MolecDyn/Molec.html. Projects like these are prime examples of how we can extend a successful program like SURP by access to CAVE(tm) technology.

In the second and third years the plan will also support access and training for the other applications previously listed on page C.2.1 . Because the CAVE(tm) is a university resource we also anticipate applications in architecture, arts and history where the emphasis will be more on using the CAVE(tm) for educational programs.

Because of the emphasis on training and the significance of the CAVE(tm) user interface being part of the instrument, we have devoted most of the cost shared funds available to supporting staff and a HCI faculty positions with regard to support and training. To demonstrate Virginia Tech's support of a staff position, the Computing Center has donated 50% cost share to an existing senior technical support staff to operate and maintain the CAVE(tm) computer hardware. The NSF Science and Technology Center on High Performance Polymers and Adhesives have committed a 50% position , at $25K, to create a CAVE(tm) HCI user interface programmer for the first year. NCSA has contributed to assisting Virginia Tech by offering technical support for the first year of the project as outlined in the attached letter. This additional position would be used to help us get our first year demonstration project in "Molecular Modeling" working, which would also serve as a research training component of the program. The proposed University Research Computing Initiative (1996-1998), when funded by the State of Virginia, has been designed to provide funds for a 3 year position which would contribute to the success of this project. This initiative was designed to be part of the proposed Advanced Communication and Information Technology Center (ACITC) program. A building, housing the ACITC is scheduled to be completed in 2000 at Virginia Tech. This research plan demonstrates Virginia Tech's commitment to create an environment where information technology such as a CAVE(tm) will become part of the university plan for computing and information technology in the 21st century.

From the same University Research Computing Initiative funds has been targeted for supporting 50% Faculty positions as well as $800,000 for equipment and software. Because the ACITC will be a state regional resource our long term plan is to include industrial access where the university will establish reasonable access fees. In a recent trip to NCSA, key administrators from Virginia Tech studied NCSA's Industrial Affiliates Program. What NCSA has done nationally we hope do regionally.

NSF Budget Form 1030 will be posted here as html table for the first, second, third years and total for all three years