CASA

Computer Augmentation for Smart Architectonics

(May 9, 1995)

Jason Leigh, Christina Vasilakis, and Craig Barnes


Description of Chris Vasilakis' Masters Thesis Show

Description of the CASA Project
VRML-ready OpenInventor-based Models
Exploring Interface Ideas in Virtual Environments
CASA Intelligent Agents
CAVE-to-CAVE in CASA
Beyond CAVE-to-CAVE: Virtual Worlds with a Legacy


Description of the CASA Project

Homes or buildings that utilize technology in order to be "smart," computer augmented, automated, "intuitive," energy efficient, or just convenient, have been in existence since the mid-1980s. The technology used has ranged from simple timers that control household appliances at designated times during the day, to computer control of office environments.

In the latter, the goal has been to hide the computers in the environment while providing support for humans in their everyday activities. There are already many instances where humans have used computers in this manner. These computers have been hidden inside everyday appliances and equipment like automobiles and VCRs. However, it is interesting to observe that the most difficult to use appliances are typically those that have the most computer-like interface (e.g., VCRs and microwave ovens), whereas the ones easiest to use are those that hide the computer interface (e.g., cars and refrigerators). Beyond the management of "creature comforts" for homes, "smart" environments can be invaluable aids in the work environment.

One environment we have been considering is the computer augmentation of the beamline control system at Argonne's Advanced Photon Source. The beamline is a resource for scientists who conduct xray crystallogrphy. The operation of the beamline hardware and software requires a complex series of specific steps that result in the gathering and analysis of gigabytes of data. Since the beamline is a time-shared facility, users operate under strict time constraints that require them to work 24 hours a day for a period of about four days. These stringent conditions, in the least, lead to user fatigue. However, fatigue can lead to user error in the handling of hardware that can result in costly damage of equipment. For example, forgetting to place a beam stop before firing the beamline can destroy a $100,000 detector. In addition, mistakes caused by fatigue can induce errors in data collection, which may eventually corrupt entire experimental results. It is these types of errors which we propose could be avoided through the implementation of computer-augmented environments.

We believe the CAVE is a fascinating possibility for use as a prototyping tool for these types of environments. As we all know, architectural walkthroughs are considered the "killer application" for virtual reality, and the CAVE is one of the better ones for this. We decided it would be interesting to take the next step in architectural walkthroughs, and put some intelligence inside the architectural models -- in essence providing a virtual-reality testbed for designing and debugging "smart" environments.

The first prototype of CASA was displayed recently at EVE4 (Electronic Visualization Event 4) in Chicago on May 9, 1995. The prototype featured a tour through a virtual "smart" home, depicting a house of the future. The prototype served as a means to experiment with a number of enabling technologies, one of which included CAVE-to-CAVE collaboration.


VRML-ready OpenInventor-based Models

The 3D models used in CASA were designed in SoftImage and modelled in the CAVE as OpenInventor objects. This came from a realization that OpenInventor was fast becoming the de-facto standard for VRML. OpenInventor was chosen over IRIS Performer because OpenInventor was a more generalized model for 3D graphics that is beginning to appear on Macintosh and PC-based microcomputers. In addition OpenInventor featured a rich set of rendering optimizations that included render caching, scene culling, and level of detail control. Recently we downloaded many of the VRML models from the WEB for display in CASA. The effects were so stunning that we decided to work with NCSA to implement traversal links into CASA so that CASA could be used to browse VRML files much like SGI's WEBSpace.

One idea we are pursing with NCSA, that was spawned from CASA was that of developing Community-VRML (C-VRML). Presently, VRML is a static model description language much as HTML is a static hypertext description language. We are interested in pursuing the possibility of using the CAVE as a browser to bring over a C-VRML document that will also provide the kind of networked support that will allow multiple C-VRML viewers to remotely connect to the server site and engage in collaborative sessions or meetings from within the virtual environment generated by the C-VRML viewer and the server's database. Currently, net-surfers access a WEB site, to simply read text, view images etc; and then they move onto another site- it is a very individual process. If we established some kind of networked C-VRML viewer, surfers could connect to a WEB site, and not only read, & see what is there, but also see who is currently there and be able to interact with them in a discussion about the material at the site. This is a lot like MUDing however we're taking MUDing to a higher level by making the entire Internet and all its WEB sites a huge MUD with diverse amounts of information and people that you normally would not find on the current dedicated MUDs.


Exploring Interface Ideas in Virtual Environments

In the CAVE, interaction is severely limited by an Ascension tracked wand which provides three pushbutton inputs and a pressure-sensitive joy-pad. CASA offered three additional input schemes: 1. voice-recogntion, 2. continuous-stroke gesture recognition, 3. the Virtual Visor and InYerFace (InYerFace is not a misspelling).

Voice Recognition

I will not describe the voice-recogmition component as it simply amounted to integrating a commercial voice-recognition system with the CAVE. This has already been successfully used in many other virtual environments.

Gesture Recognition

The continuous-stroke gesture recognition scheme allows the user to train the system to recognize a sequence of continuous wand movements. A press of the wand button begins the gesture recognition. The user than moves the wand to draw a gesture. On conclusion of the sequence the wand button is released. The computer then examines its database to determine which gesture was just executed. This technique was borrowed from hand-writing recognition systems like Graffitti and Unistroke developed for the Apple Newton Message Pad. Using this scheme we were able to train the system to recognize all 26 characters of the alphabet.

Virtual Visor

The Virtual Visor is a virtual display device and input device for the CAVE. It simulates a heads-up display in the CAVE on which status information for a particular set of simulation parameters can be displayed as numbers or dials. The InYerFace provides the visor with the additional functionality of being an input device that is controlled using the user's head orientation. When in InYerFace "mode" the visor freezes in space and a targeting cursor appears in front of the user. When the user moves his/her head the targeting cursor instantaneously follows. To make a selection the user aims the cursor at the selectable visor item (simply by looking at it) and presses a button on the wand.

The reason for using this kind of paradigm is that in contrast to everyday life where tactile feedback allows you to operate things without necessarily looking at them, images in virtual environments still have no tactile properties. Hence when you make selections with a menu you always have to look at the target first and then use your wand or glove to make the selection. The task involves three steps, 1. seeing the target, 2. aiming the wand or glove at the target and 3. making a selection. With the InYerFace, step 2 is eliminated because once the user acquires the target visually, he/she is already ready to make the selection. One additional benefit of the InYerFace comes when it is used in Head-Mounted Display, Fish-Tank VR systems, and the ImmersaDesk(tm). In such systems there is a common problem of fatigue that occurs due to prolonged raising of the user's arm to make selections. The InYerFace can be used to greatly reduce this problem by reducing the number of operations that require arm movements.


CASA Intelligent Agents

In our prototype of a "smart" house, the lights are driven by sensors that are placed in specific locations in the house that can be triggered when they are approached. Instead of hardwiring this kind of behaviour we developed a notion common to intelligent agent systems. That is, the "smarts" were programmable by a script that can be used to declare the sensors, the controllers and the controlled objects. For example in the "smart" house, each room had sensors which, based on certain conditions, would trigger one or more lights to turn on. The class of sensors and controlled objects is extendable by declaring subclasses of base sensor, and control objects. The logic for the controllers are programmable via a scripting language.

CAVE-to-CAVE in CASA

The CAVE-to-CAVE project is a collaborative effort between the Electronic Visualization Lab, Argonne National Lab, and the National Center for Super-computing Applications, as a subproject of providing networked collaborative virtual environments for the I-Way.

The Information Wide Area Year (I-WAY) is a proposed experimental high-performance network linking dozens of the country's fastest computers and advanced visualization environments. This network will be based on Asynchronous Transfer Mode (ATM) technology, an emerging standard for advanced telecommunications networks. The network will support both TCP/IP over ATM and direct ATM oriented protocols. This network will provide the wide-area high-performance backbone for various experimental networking activities at SuperComputing '95. (For more information see: http://www.anl.gov)

The entire CAVE-to-CAVE effort is currently divided into a number of efforts. EVL has had a long standing history and considerable expertise in technology transfer. Consequently EVL's role in the CAVE-to-CAVE effort has been to determine the needs of its potential users and to communicate these needs to Argonne National Laboratory, who are developing the networking software to enable CAVE-to-CAVE communications. EVL is primarily interested in researching new interaction techniques and applications of distributed collaborative virtual environments.

CASA's CAVE-to-CAVE component allowed multiple networked participants (running the CAVE or the CAVE simulator) to explore the same CASA space. Each participant could choose avatars that they designed themselves using various 3D modeling packages. The avatars consisted of a head, a body and a hand. These components derived their orientation and position from the CAVE head-tracker (for avatar head orientation), CAVE world-space navigation (for body position) and CAVE wand-tracker (for avatar hand orientation and position). These separate components provided users with greater expression with their avatars as it allowed gestures such as nodding of the head and waving of the hand.

The CAVE-to-CAVE communications was provided by a communications library developed at EVL, called SpiffNet. SpiffNet is a C++ based library that provided near-transparent access to an internetworked centralized database server. When programming with SpiffNet users worked with data objects as much as they do with objects in the CAVE's shared memory. SpiffNet provides a number of base data types like float and int called nfloats and nints (for network-floats and network-ints), which when instantiated allowed the users to treat them exactly as regular ints and floats in C. The fact that they are actually networked variables that may be shared by other CAVEs is hidden from the programmer. The underlying client manager determines the most optimal way of broadcasting changes in these variables. The central server maintains a consistent database schema of all the networked clients and broadcasts these changes to all clients that need the data, and at data rates compatible with the clients. This prevents slow clients from being inundated with data from the server.

By using a general networking system, rather than a CAVE-specific networking library, we were able to connect other non-CAVE clients to the CAVE. For example we were able to digitize voice on a remote SGI Indy and send amplitude information to the CAVE, that could be mapped onto an avatar's head to simulate lip-synching. This lip-synching might be a lower bandwidth solution to transmitting facial images across conjested network lines.


Beyond CAVE-to-CAVE: Virtual Worlds with a Legacy

Finally, an area we are greatly interested in exploring in collaborative virtual environments is the notion of never-ending virtual worlds. That is, virtual worlds that continue to exist and interact with either other participants or intelligent agents, even when a local client has closed its connection to the virtual world. Just as real life continues whether we wish to participate in it or not, virtual worlds should. This allows users to enter a virtual world, summon intelligent agents or other participants to perform tasks that may take some time to complete, and then go away to do other things. If the user wishes to check on the status of the agent, the user can simply re-enter the virtual world.