Dick Pountain - 28/02/95 11:41 - IDEALOG 7
Wherever I look nowadays there's someone playing Doom. In
fact I suspect that half the PCs sold last Christmas were
really bought just for that purpose, and it set me to
pondering why this game has become so extraordinarily
popular. It's not just the scenario of blowing people to
bits with an assortment of hellish hardware as plenty of
other games that offer that, with even more flying
ketchup. What really grabs people about Doom is the
stunningly executed implementation of 'virtual reality',
which even on a humble VGA PC screen can convey the
impression of motion well enough to make you travel sick.
The truth is that those murky Doom corridors are more
impressive than most of the demonstrations of immersive
(ie. using a headset) virtual reality that I've seen -
Doom's level of detail contrasts so strongly with the bare
unshaded polygons that are the best most low-end immersive
VR systems can manage.
Virtual Reality has caught the popular imagination in a
way that computing matters seldom do, which is hardly
surprising given that most citizens perceive computing as
a grey (or perhaps biege?) profession that concerns itself
with numbers, gas bills and things with giggle-making
names like floppies and bits. To most people computers
only become interesting when they show images, and even
then those images often disappoint - TV and the cinema
have visually conditioned us by now to expect
photorealistic, moving images which today's PCs can't
deliver. We in the computing profession rarely appreciate
the depth of this disappointment because we know too much
about hardware limitations; like Dr. Johnson commenting on
the dog 'walking on its hinder legs', we tend to be
impressed that it can be done at all, while the layperson
just sees that it is 'not done well'.
The arrival of Virtual Reality signals that computers can
finally deliver convincing images of the world, which will
eventually make them stunningly powerful tools for the
visual arts. VR promises to be Science's answer to the
ancient craft of Magic, enabling us to make fantasy flesh
and to change the (virtual) world by thought alone. In a few
years you may be able to climb a virtual mountain on Mars
or walk through ancient Rome.
Inevitably, some folk now go overboard and prophesy that we
will eventually build a virtual world big enough to live
our whole lives in. How likely is that to happen? Pretty
unlikely actually. You can get a handle on the size of the
problem by extrapolating from current VR systems and
making reasonable assumptions about progress in hardware
design, and the sums don't look too good. Merely
rendering photorealistic moving images is a tough enough
problem. To sustain believable animation you need to
display at least 10 frames per second, and for true
photorealism each frame would need to be made up of nearly
80 million polygons, requiring the graphics engine to draw
approaching a trillion polygons every second; Silicon
Graphics current top of the range RealityEngine2 can
manage 2 million polygons/second, so some way to go there.
Now you know why Doom uses those big chunky pixels.
These figures apply to 'pure' virtual reality systems
where there are no stored pictures at all, but only a
database of descriptions of the virtual world's geometry
from which the latest viewpoint is computed for every
frame. A couple of years ago I was playing with a
space-walk simulator developed by the British VR firm
Division for the European Space Agency. I asked Phil
Atkins, Division's techno-wizard, why the smoke that
erupted from my thruster pack looked more like a grey
carrot, and he just said, "Think about it, Dick". So I
thought, and the light dawned; to have realistically
behaving smoke would have meant keeping a separate
database entry and geometry calculations for EACH SMOKE
PARTICLE.
However there are ways of 'cheating' that can cut the task
down a lot, which is how Doom manages to look so good on a
humble 486. One is texture-mapping, where you store some
bitmapped pictures of common substances (like brick, stone
or wood) so that when you want to depict, say, a door you
just draw a rectangle and then distort your wood bitmap
into it like wallpaper; in this world all wood is the same
wood. Similarly you can use object instantiation to copy a
thousand bricks from a single template, whereas in the
real world you've no option but to dig that clay.
To see how much difference these tricks can make, I asked
Phil Atkins to talk me though some back-of-envelope
calculations over the phone. We decided to 'cost' a
photorealistic VR representation of one square mile of
Central London. We'd squander around 30 polygons on each
building less than 30 yards away and just 6 polygons for
more distant buildings. We'd texture-map scanned
photographs of the actual buildings on to these polygon
skeletons, while perspective-transforming the photos in
realtime through Division's VPX PixelPlane engine which
can process several millions of polygons per second. It
turned out we'd need roughly 60 Gigabytes of texture
memory and a few hundred megabytes of geometry storage.
That's almost thinkable today (at a colossal price) but it
would still only give us outside street views; to be able
to go inside each building might multiply that by hundreds
of times. And that's only vision; add stereo sound, and
solidity or feel data (which is still in its infancy) and
you begin to see the size of the problem. As for poor old
smell, no-one even knows where to start.
No comments:
Post a Comment