jraymond@BBN.COM (Jayson Raymond) (06/25/91)
Our esteemed moderator writes: >As moderator, I would like to encourage the discussants of engines >and drivers to "come on back!" The topic of machinery is appropriate. >We need better computing hardward to make virtual worlds that are useful. >In the last two days, the HIT Lab has heard impressive presentations from Herb >Taylor of Sarnoff Research Center, regarding the Princeton Engine; and >people at BBN Applied Simulations regarding their CIGs (correct acro- >nym?) rapid simulation system. It was clear to us that the emergence of >this technology may change the way we make virtual worlds, especially as >this gear gets smaller, cheaper, and faster. Okay, I'll take this opportunity to plug Computer Image Generators (and "CIG" or "IG" are correct acronyms) in general and their applications to VR. These systems are specifically designed to provide three dimensional, polygonal, textured, virtual worlds that are typically unobtainable at adequate performance levels by general purpose computers. (As you read the following, please bear the following in mind. (- First, I am speaking from my own personal interest in VR and in no way represent BBN or other CIG companies mentioned herein. (- And second, although I have focused on the technology and not the applications and the typical financer of such technology, the fact remains that those controversial defense dollars have made much of this technology possible, and continues to drive the prices down and system performance levels up, making references to defense unavoidable.) A little history ---------------- Previously the high cost associated with designing specialized systems with the performance required to produce scenes that suspend disbelief resulted in simulators that were (and many still are) quite costly (> $1 M). The only applications that could afford this type of budget were aircraft trainers, with little-to-no interaction with other aircraft. As technology advanced, the cost of producing a scene that was a "60% solution" (320 x 128) had reduced to the point that made a virtual world comprised of hundreds of networked participants attainable. This allowed team training to occur, as opposed to the individual training focus of previously. This concept was proven in the Darpa funded Simnet (SIMulator NETwork) program, with 320 worldwide networked ground vehicle and aircraft simulators currently operational. Due to the success of the Simnet concept for training and concepts prototyping, many countries worldwide, as well as the US are either investigating or in the process of implementing full scale networked vehicle simulators. Implications to the VR community -------------------------------- The result is that many companies are scrambling to design Computer Image Generators that can be produced on a large scale, thereby bringing the costs of these systems into affordable ranges for VR laboratories (around $250k, give or take). And these CIG's are equipped with features, not available on general purpose systems. CIG's often are equipped with multiple channels. So instead of utilizing two Silicon Graphics workstations per user (one for each eye, as in VPL's system), one system can provide views for both eyes, or even multiple participants. For example, Silicon Graphics' new Skywriter provides 2 channels at resolutions greater then 512 lines, while BBN's GT102 can provide 16 channels (though at lower, Eyephone resolutions). Also, most CIG's are frame driven, as opposed to the run-to-complete processing of typical general purpose systems. Regardless of the scene complexity, a new image will be provided every 15th of a second (or whatever time slice chosen for the application). This eliminates the disorienting, "clunky" feel associated with images that are produced on "run-to-complete" systems. Of course if the image wasn't completely done being generated by the time the frame reset came around, an incomplete image was presented. This is deemed an acceptable degradation in systems that process front to back, thereby ensuring that foreground objects take priority. Also of benefit are the now emerging standards. Database standards allow database's to be shared and hosted on other hardware platforms. And networking standards are being established so that anyone can "plug-n-play" into existing networks of simulators. Current spec's for future networked simulations are pushing the polygon count and resolutions to new heights in the low cost simulation market. And texturing requirements are specifying that there be no performance degradation when using textured polygons. (May there be a texture for every polygon, and a chicken in every garage..). So in short, if you are considering buying or building a VR system, depending upon your application, a Computer Image Generator could quite possibly be the best choice. Afterall, why use 2 workstations for one user when for a very similar cost, one CIG and it's advantages could drive 4 users equipped with Eyephones? And finally, if you are interested in learning more about CIG's, the following companies and systems can provide a starting point. Evans and Sutherland's ESig 3000, GE's PT2000, Star Technologies Graphicon 2000, Silicon Graphics SkyWriter, Paragon, Ivex, BBN Advanced Simulation's GT100, and others. The above is from memory, so I apologize for any errors, Mea Culpa. -- Jayson ------------- Jayson Raymond jraymond@bbn.com BBN Advanced Simulation 14100 SE 36th St. Bellevue, Wa. 98007
galt@es.com (Greg Alt - Perp) (06/26/91)
In article <1991Jun25.154814.27460@milton.u.washington.edu>, jraymond@BBN.COM (Jayson Raymond) writes: |> CIG's often are equipped with multiple channels. So instead of |> utilizing two Silicon Graphics workstations per user (one for each eye, |> as in VPL's system), one system can provide views for both eyes, or even |> multiple participants. For example, Silicon Graphics' new Skywriter |> provides 2 channels at resolutions greater then 512 lines, while BBN's |> GT102 can provide 16 channels (though at lower, Eyephone resolutions). Just thought I'd mention that Evans & Sutherland's ESV can generate stereo images even on the lower end systems. The high end (~$90k) can do over 100,000 (10x10 pixel) shaded polygons/second, as well as over 1,000,000 (4 pixel) vectors/second. The low end (~$30k) can do around 20,000 polygons/sec ond. People keep quoting prices for machines > $200,000, but I would think that for around $50,000 you could get a high-quality VR system using an ESV, a data-glove, and eyephones. The graphics programming would be done using Phigs+. A while back, someone from Florida (Univ. of Florida?) said that they were using an ESV for VR. Greg --------- I'm not an official spokesperson for anyone, so it's not my fault.