LOW COST VISUAL SYSTEMS COME OF AGE
Gregory F. Gustin
Paragon Graphics, Inc.
Visual simulation has been, without a doubt, the most elusive segment in the evolution of cost-effective simulator training devices. Advances in the state-of-the-art have come at a phenomenal price. During the mid-1980's, basic CIG systems have cost upwards from 1.5 million dollars, and the most sophisticated systems have cost over 5 million dollars. Now, in 1990 CIG visual systems can finally be acquired at very significantly reduced expense, about one-tenth that of the previously established price levels. Additionally, these new low cost CIG systems include many standard capabilities that were only recently very expensive options on the most advanced systems: features such as multiple eyepoints, independently moving models, anti-aliasing and transparency.
The exotic flight simulators used today by the military and by commercial air carriers incorporate multi-million dollar computer image generation (CIG) visual systems. The visual simulation industry has progressed in less than ten years from the delivery of night only systems, to dusk/night, and finally to full color daylight displays. Scene content - whether defined in faces, edges or polygons - has become a primary standard for measuring system performance, and it too has evolved steadily. In the 1970's, 300-500 faces was the accepted norm, while today's most advanced systems are measured by the thousands of faces.
As with any added value, there always remains a price to pay. And, pay we have. Literally millions of dollars have been required to purchase and integrate state-of-the-art realtime CIG systems. In the 1970's, the CIG system became the most expensive element of the flight simulator, and it has remained so ever since.
Yet, every educational technologist knows there is a multitude of training applications that could benefit immeasurably from employing realtime CIG devices - if only the cost were in reasonable proportion to the value of the skill being trained. In aviation, the list includes providing a greater number of visually equipped cockpit simulators and target image generators as well as re-equipping existing cockpit simulators with advanced color day CIG systems to replace antiquated dusk/night devices. For ships and land vehicles of all sorts there are numerous training missions suitable for CIG visual systems...if the price were low enough.
The past three years have seen the dawning of a new era in the CIG marketplace, and this new era has been marked by revolutionary changes in what CIG systems cost and in what they are used for. Today CIG systems exist - ready for immediate delivery in multiple quantities - that offer the performance equivalent to the state-of-the-art systems of just a few short years ago. But, these new systems are priced from only $50,000 to $250,000. Additionally, these new systems are offering special features that were literally unavailable three to five years ago, and offering many of these features as standard equipment.
How we got from where we were in the 1970's and early 1980's to the revolutionary technology of today is certainly worthy of examination, and this paper includes a cursory review of CIG history. The paramount message of this paper, however, is that a radically lowered price structure is beginning to open new vistas of opportunity for employing realtime CIG visual systems as training tools. And, "rapidly" is the key word; these new low cost CIG systems have been fielded in very significant quantities during the past three years. Further, just as the jet fighter replaced the propeller aircraft of World War II in less than a decade, these low cost configuration CIGs will come to dominate the visual system marketplace by the end of this decade.
Where We Were: A Brief History
The first CIG device suitable for training began its evolution at General Electric in the early 1960's. GE undertook this effort on behalf of NASA to support vehicle docking training for the Gemini program. The earliest versions of the device displayed little more than a checkered earth surface and simple vehicle replication. Development of the basic technology continued during the next ten years, and by 1972 GE was able to deliver to the Navy the first full color raster CIG device for use as a visual flight simulator in military training applications. Called the Advanced Development Model (ADM) or the 2F90, it was employed to measure the effectiveness of computer generated imagery for pilot training.
Although this prototype system was found to provide inadequate velocity and attitude cues for lineups and landings on a simulated aircraft carrier, the potential of CIG technology was not itself held in question. Rather, as GE continued to evolve its system's capabilities, other major companies undertook development of similar devices.
McDonnell Douglas introduced the first CIG system to the commercial airline industry in 1971 with the installation of a Vital II system on a Pacific Southwest Airlines (PSA) 737 flight simulator. This prototype system demonstrated what could be achieved by calligraphic techniques (lines and light points), and it impressed the FAA sufficiently to gain approval for use in commercial pilot training as a substitute for flight hours, and even as a substitute for airborne currency (line) checks.
By 1982, four companies had become recognized as primary suppliers of CIG simulation systems: General Electric, McDonnell Douglas, Evans and Sutherland/Rediffusion and Singer-Link. During the early 1980's, CIG capabilities increased profoundly, but the prices remained very high. Color day displays, introduced in the late 1970's for several million dollars, continued to cost nearly an additional million dollars per channel over dusk/night systems. Anti-aliasing, a popular enhancement demanded by many customers to smooth the "stair-stepping" found in raster systems, as well as moving models independent of dynamic background scenes, could now be provided with the most sophisticated systems, but only for a few very well funded customers.
Virtually all of these and other significant advances were introduced by the established "big four" companies. In the mid-1980's these giants had continued to methodically advance the state-of-the-art, while carefully balancing their selling prices against one another's equivalent products.
A small number of additional aerospace companies jumped on the CIG bandwagon in the 1980's. Tector (from England) was the first to introduce a "low cost" full color raster CIG system. The Tector system offered a 3 channel, 4 window configuration with displays for less than one million dollars. This extraordinary price was, however, not enough to compensate for two serious design limitations: first, the inability to display a three-dimensional object (buildings, hills, etc.); second, a significant utilization of analog electronics which inhibited the device from displaying stable images demanded by most customers.
Others soon followed: Sogitec and Thompson-CSF from France and JRC (Hitachi) in Japan. While these companies each succeeded in establishing their presence in the marketplace, none of them have achieved a track record of sales that would warrant including them - on a sales volume basis - in the same grouping with the "big four". These newer CIG manufacturers have all produced products that are commendable, providing viable applications of CIG that were literally unobtainable a mere five years previous. Yet, each of their systems came hobbled with at least one compromise: the update rate was too slow, or the scene content was too restricted, or the integrated system was prohibitively expensive.
The Winds of Change Begin to Blow
The mid 1980's saw another change in the marketplace. A handful of new companies previously not involved in simulation - and in some cases not even in existence - began to tout their capabilities at supplying CIG systems. Although there are certainly others, four of the most prominent exemplify the new trend and are discussed here.
Silicon Graphics found its way into visual simulation as an offshoot endeavor growing out of its highly successful track record at producing innovative graphics engines for the CAD/CAM industry. Indeed, the Silicon Graphics Iris workstation was readily adaptable to producing CIG scenes in color and with sufficient content to produce reasonable detail for out-the-window simulator displays. With an average selling price of $100,000 to $200,000, it was an appealing choice for training applications desperate for interactive CIG visuals but where there was little hope of obtaining two million dollars in funding for a standard CIG system. The compromise: a maximum update rate of 30hz for low detail scenes and less than 10hz for scenes containing more than 200 full screen polygons. The IRIS family of graphics workstations, powerful as they are, should not be categorized with devices that we describe as realtime CIG systems.
In 1982, the most innovative new CIG system was the creation of Trillium Corporation, the first "newcomer" to design and deliver a realtime CIG. Dr. Ron Swallow left ATS, a division of the Austin Corporation, to form his own company, Image Dynamics, Inc. In less than 15 months, Dr. Swallow and his small team of scientists were able to design and develop a system that produced color displays containing more than 1000 smooth shaded polygons. The small company quickly realized that far more than a name change and capital infusion from Wall Street was necessary to establish the Trillium product in the visual simulation industry. Low display resolution (640x480) and lack of anti-aliasing prevented the Trillium system from being readily accepted. Although the price of less than $150,000 per channel was attractive enough to some, Trillium could not evidence the capability to evolve from a design-centered operation into a full-fledged production organization.
Another corporation, GTI, also recognized the potential of the realtime CIG industry and established a graphics division. Their premier product, the Poly 2000, was not as powerful as the Trillium machine, yet it achieved a greater market success. GTI, using the strength of its parent corporation, was able to develop a production facility, thereby assuring its potential customers that product, once purchased, was indeed likely to be delivered. After some years of development efforts, the Poly 2000e (e = enhanced) was demonstrated. This system had considerably more features than the original Poly 2000 (e.g., smooth shading and anti-aliasing) but was still plagued by low resolution (i.e. 640x480). GTI also lacked the internal organization that could successfully integrate with the industry.
IVEX Corporation, founded in 1983, originally conducted research and development in the field of interactive video games. The research team, a group of computer graphics engineers from Georgia Tech Research Institute, recognized the adaptability of their product for utilization in realtime simulation and training. In late 1986, the IVEX group was able to prove, to the surprise of industry observers, that videodisc technology could indeed be used as an integral subsystem for realtime visual simulation. It was not the laser disc itself that was at issue; others had successfully employed similar approaches. Honeywell's CGSI and Vought Visual Technology had both produced videodisc-based systems, but each of these systems cost several million dollars. The IVEX system, on the other hand, was offered for less than $200,000. The "strength" of the IVEX system is its ability to portray extremely robust scenes to the viewer which have proven to be quite acceptable in meeting the ab initio training requirements. Initially, the scenes were limited to a two-dimensional world representation of the database, that is, no hills, buildings, or vehicles. However, in 1989, IVEX demonstrated the ability to mix 3D polygonal objects with their own scene generator. This additional capability seemed at first glance, to resolve one of IVEX's serious limitations, that is, 2D database restrictions, yet, as of this writing, the excellent fog and weather effects in the IVEX system are not applicable to the 3D models (terrain or cultural features). Low display resolution, as well expensive and long lead times for custom databases, inhibit the widespread acceptance for applications requiring more than take-off/landing training.
In November of 1985, a company called hi-tech MARKETING CORPORATION (HTM) displayed, for the first time, a CIG product called "Black Box I." Priced at $35,000, key features included a 30hz update rate, color day displays and an on-screen data base display capability of 300 polygons. In March 1986, the Black Box I system was demonstrated in a three channel integrated format displaying a 50ox160o FOV. And, in August of 1986, an enhanced Black Box I was introduced. The Black Box I was the first to use a 32-bit floating point data structure rather than the typical 16-bit integer format. Other enhancements included multiple moving models and realtime data base management techniques.
What were the compromises associated with HTM's Black Box? Well, technically, there were none. (If 300 polygons doesn't seem to be very much scene content, remember that almost all of the CIG systems sold in the 1970's could produce only 300-400 polygons and that most of these systems had limited color capability). In the customer's eye, the perceived compromise was taking the risk of doing business with a new, small company that had never manufactured anything before, let alone something as exotic as a CIG visual simulation system.
Where Are We Right Now?
Looking forward, three paths lie before the CIG visual system industry, and there are travelers to be found on each of them. For discussion, these paths may be labeled: 1) the best at any price; 2) lower cost; and 3) low cost. On the first path we find the manufacturers and their customers who will accept only the newest and most elaborate systems supplied by the world's most credible sources. It has been proven possible to convince the military and the major air carriers that only a "Rolls Royce" will do, and some CIG manufacturers will keep delivering their luxury models for some years to come, evidenced by GE's CompuScene V & VI, as well as the E&S family of new high-end systems, ESIG 1000, 3000, and 4000.
"Lower cost" CIG's have become an option only recently with the introduction of various less expensive systems offered by major manufacturers. Singer/Link (recently purchased by Thompson CSF) has its IMAGES system, quoting prices that are half as much as the standard DIG series. Evans and Sutherland has, in addition to its SPX systems, announced a texture upgrade capability for its NovaView series (now offered through a Hughes division formerly Rediffusion). General Electric has introduced the PT2000 to supplement its CompuScene series.
These systems, and perhaps others soon to be announced, may be hundreds of thousands of dollars less than similar systems sold three to five years ago. But, relative to paying 3 to 5 million dollars for a new commuter aircraft, does this represent a low price? For a training center with an annual operating budget of 1 million dollars, is this truly a low price? Relative to charging a flight student an additional fee to help amortize the cot of the school's visual flight simulator, is this a low price? Yes, lower priced CIG's have arrived, but it is decidedly not correct to place these offerings from the "Big Four" in the same category as truly low cost systems.
Path three - the low cost systems - are needed, and the need is now. They have been needed ever since we discovered how to build a cockpit simulator of reasonable fidelity for less than one million dollars. A good number of manufacturers can supply an effective single or twin engine aircraft simulator for about half a million dollars. The general aviation aircraft manufacturers are delivering basic training aircraft for under one hundred thousand dollars. Therefore, with regard to CIG visual systems, it seems reasonable that "low cost" be defined to mean ready-for-training systems that cost less than $100,000 per channel.
Without low cost systems, the need for trained operators will remain insatiable. Without low cost systems, we will suffer not only from too few trained operators, but those trained will not be of the caliber that we are accustomed to today. Yet, the conventional wisdom has been that low cost systems could not be delivered with sufficient functionality to provide worthwhile training. Despite conventional wisdom, low cost visual systems have come of age. In the past three years, Paragon Graphics, Inc. has provided irrefutable proof that a fully-featured CIG visual simulation system can be designed, manufactured, integrated and operated for less than one-tenth the previously accepted price for systems with equivalent performance.
Where Are We Going?
Paragon Graphics, Inc. has established a position of credibility within the industry by successfully delivering and integrating its PARAGON RVS (Realtime Visual System) on to a number of customers in both the military and commercial sectors. Paragon was founded in November 1985 by Carlos A. Sampayo and Mr. Gregory F. Gustin. Previously, Mr. Sampayo was a highly respected independent researcher and developer of advanced CIG systems. He was the senior design consultant for the naval Training Systems Center's Visual Technology Research Simulator (VTRS) facility. His accomplishments there included simulator visual systems design and integration, 3-D data base development and testing, and the design and programming for state-of-the-art array and bit slice processors. He also invented and implemented the VTRS POCCIG, and he created the GPX system, a 3-D polygon raster rendering package. Mr. Sampayo, prior to his NTSC involvements, was a visual systems engineer for the Simulation and Control Systems Department of the General Electric Company. There he contributed to the research, development and implementation of a series of modifications to the NTSC VTRS CompuScene visual system. Mr Gustin's experiences include a tour of duty as a U.S. Marine Corps fighter pilot and aerospace systems engineer. He has earned special recognition for his revolutionary applications of computer graphics to projects under his direction at the Pentagon, McDonnell Douglas Corporation and the Northrop Corporation, and was responsible for the Black Box I CIG produced by HTM.
In less than twelve months, Paragon Graphics designed, developed and demonstrated the Paragon I RVS. The PG-I was the first truly low cost CIG which could duplicate all the key performance features previously available exclusively on multi-million dollar systems such as 1000 line resolution, anti-aliasing and a 60hz update rate. Although low cost, the PG-I system is capable of presenting robust out-the-window scenes (Fig. 1), dynamic background scenes containing multiple moving objects (Fig. 2), and high detail target images (Fig. 3). PG-II provides vastly increased performance capabilities over PG-I, including an increase in realtime scene content of 10,000 versus 1,000 displayed polygons. PG-II is also capable of supporting other options such as smooth shading, Z-buffer hidden surface calculation and dynamic distortion correction.
Both systems include all the standard features associated with modern CIG systems: sub-pixel anti-aliasing, light points, weather effects, horizon glow, fade level detail, sun shading, transparent polygons, self-illuminating polygons, graceful system degradation, landing lights and data base management. PG-III, already demonstrated in prototype configuration, offers dynamic texture mapping. Each photo map can support up to 250,000 pixels, and 64 maps can be utilized per frame. The arrival of PG-III has made the performance of the most sophisticated systems affordable to nearly all users.
Since the inception of simulation as a training media by Mr. Ed Link, the trend has always been toward the goal of achieving greater realism. For years, industry accepted the higher prices that accompanied higher fidelity. Other than replicating sustained "G" cues, simulation can now replicate the vehicle in all aspects - especially when one views the visual scenes utilizing photographic photo mapping.
The question now being raised is when is the level of fidelity too high? When are we presenting the trainee with simulation far more detailed than is required to meed training objectives? For example, leaves on the trees certainly may be necessary for nap-of-the-earth training missions, yet simple tree shapes are sufficient for most scenarios. But, can we stop without any one of us ever needing moving leaves, or moving bugs on the moving leaves?
Obviously, when training device cost exceeds the value of the skill being trained, the price is too high. When the cost exceeds that of training on the device itself, the price may also be too high unless additional valuable benefits are gained such as avoiding catastrophic accidents and reducing training time. Certainly everyone has come to recognize that the value of simulation is greater than simple monetary savings over real vehicle operation: we all extol the virtues of practicing procedures safely (e.g., fires in flight, wheels-up landings) and tactical engagements (e.g., expending munitions) and enhanced instructional environments (e.g., pause, freeze, replay).
The question is, has our industry comprehensively determined which system features are necessary as critical training cues and which features are simply gold plating? Can we all accept that it is not appropriate to put leaves on the trees and waves on the water when teaching students the basics of flight? Can we all agree that it is not necessary to show various cloud strata to trainees during their first attempts at learning landing approach procedures? And, can we sell our determinations to those who buy and to those who use our CIG devices?
There is no doubt that various visual effects greatly enhance the realism of the presented scene. There is also no doubt that these enhancements to scene realism improve pilot acceptance, and pilot acceptance is one of the most important features of any device. But, at what price... acceptable... unreasonable... outrageous?
The naive answer would be to make the features of the multi-million dollar CIG systems (i.e., photo texture mapping) available for the same price as the "low cost" polygonal machines. Concurrent with a reduction in features pricing must be an accompanying increase in system reliability. With the delivery of potentially thousands of realtime visual systems to the training community, 50 hours or even 500 hours MTBF, once thought impossible to achieve, would still be a fatal shortcoming. Systems reliability will have to approach 20,000 hours, that is, be capable of unattended operation. Any failure that does occur must be diagnosed to a board level with R&R capability of less than 5 minutes.
Difficult goals? Not necessarily. The first PARAGON RVS delivered has accumulated over 9000 hours of operation without a single maintenance hour. The availability of VLSI, gate array technology and soon VHSIC will bring these numbers to an even higher level of performance reliability.
Customers have become more and more unwilling to purchase databases for prices that are higher than prices for the actual hardware. Data base modelers have been able to create aircraft models with less than one week training on PARAGON systems. The PARAGON system power, when coupled with its delivered efficient software modeling packages reduces "at work" prices to five dollars per displayable polygon. Other requirements begging for change by the customer include ease of installation, as well as ease of use. PARAGON RVS systems have been installed and placed in operation in less than one hour after delivery.
Performance. Price. Availability. Reliability. The systems of the future are here today. Tomorrow will bring even greater realism at even lower prices--systems without compromise.
Presented at the Image IV Conference
Phoenix, Arizona, 23-26 june 1987