From: dstamp@watserv1.waterloo.edu (Dave Stampe-Psy+Eng) Subject: Re: half silvered lenses (was Re: Direct Neural Input (Was Re: Date: Sat, 9 Nov 1991 05:59:51 GMT Message-ID: <1991Nov9.055951.23810@watserv1.waterloo.edu> Organization: University of Waterloo craig@utcs.utoronto.ca (Craig Hubley) writes: >Would require small CCD video cameras of the kind used on David Letterman and >dropped into sinks, strapped onto monkeys, robot arms, pencils, etc... >cost about $11k each last I looked, and dropping fast. Of course then you >blow the resolution of reality, but it's a good way to prototype a system. > I recently finished a specialty head tracking system using these. The cost of *industrial* quality minicams (1.5 cm diameter, 10 cm long with cable plugged in) is about $3000 US. But there's no reason you can't use an element from another source (i.e. scrapped camcorder) for less than $500. These give you color, but you lose some resolution and can't see infrared LEDs (internal color filters kill them). But you might use a UV light and patches on a dark background, using the color info for point ID. And of course, you can always get a CCD sensor and build around it.... >>There HAS to be magnification, unless the display element (CRT, LCD panel, >>etc.) is as big as the mirror. You are increasing the apparent size of the >>display to put it at an apparent distance of 100 cm or more, while increasing >>the field of view. > >I must be missing something... how close was all this to the eye ? And how >small can you make LCD panels or CRTs ? > I assumed you were talking about viewfinder CRT's (1/2" dia. screen). Of course, larger (2" or so) CRTs will have less problems. Better assume that (without mirrors) you're at least 1" from the eye with your resolving lens, and another inch from the lens to the display. Need at least a 2" display for adequate FOV... >>In optical systems like this, motions of the viewpoint (the pupils of the >>eyes) in relation to the mirror will cause shifts in image position much >>as if the display element moved in relation to the mirror. The pupils of >>eyes move through about 12 mm during normal eye movements, but this can >>be compensated for in the optical design (just have the viewpoint be at >>the eye's center of rotation, 12 mm behind the cornea). However, shifts > >OK, but how does this work for odd-shaped (e.g. myopic) eyeballs ? We >are talking about fully 60% of the world's population here... > Not really a problem, as small shifts caused by eye movements are OK, but drift from other sources is noticeable. Effects from the tiny decentering caused by myopia can be fixed by changing the eye-to-display spacing. >>(rotational inertia) of the headmount. Movements of facial muscles are >>a problem even if the headmount is tightly fastened to the head. > >So you need this to be light and you lack a fixed point of connection >unless people want holes drilled in their skulls... maybe the headmount >needs to compensate for this a la Steadicam, either physically or by >telling the renderer where it is going to bounce in the next 1/60th second. It's not THAT bad, you just have to design for it. We got a fairly stable headmount with a welder's headband ($10) and aluminum frame. You have to be careful where it sits on the head to prevent effects from facial motions. > >>Perhaps I'm overrating the problem here, but I do beleive that this problem >>will limit the magnification usable for the mirror, setting a minimum size >>for the CRT or display element. > >How big ? I'd say that a 2" display is the minimum, but this has to be tried. Also, if your optical system distorts the image and has uneven focus, (and most present VR optical systems do), nobody will care about a little movement, they'll be too busy coping with the other problems. >>Time delays ARE a problem, even in complete VR worlds, including represent- >>ations of the user's hands, etc. The reason is that the human motor system >>does NOT use negative feedback directly, due to long neural delays (>100mS). >>Instead, it relies on Kalman-type filters and task-specific motor programs >>learned through practice. Any large change in motion-to-visual feedback >>delays requires a recalibration of the system at best, and relearning of >>fine control for a task at worst. To say nothing of motion sickness > >Are these constant for an individual ? Can it be calibrated for one person ? Problem is, human systems LEARN, so the system changes with time. Also, we don't want any adverse effects that limit use of the system. Of course, that's an ideal, so we must find out what effects small delays have, and what the best tradeoff is. Calibration is difficult, because we have no matching input to comare output against. > >>brought on by changing head-motion-to-scene-motion delays, which has NOT >>been covered well in VR literature, but exists nonetheless. > >Any references to these issues would be appreciated. > >>virtual *objects*, while letting the user see his hands, the computer >>keyboard, etc. This solves a LOT of the psychophysical and kinesthetic >>problems. But the time delay problem kills it, for now at least. (sigh). > >I don't understand why. Seems to me that the answer is practice, practice, >practice, which is absolutely justifiable if you are going to provide such >a useful device. I don't find it all that easy to walk around in VR >anyway... The bad part of the "objects only" interface is the delay between hand and object movements. So the user would tend to move objects slowly (to prevent his hand from passing thru them). Might work out OK, except that real-world-to-VR is a difficult problem-- say if the displays shifted slightly on the user's head, and everything was off by a few inches forever afterwards! >From what I've seen so far, the effect of visual feedback delay is much slower movements on the user's part. Also, there are overshoots and "hunting" in movements. This is in addition to the problems caused by lack of tactile and force feedback. Motor learning is a *lot* harder under visual-only conditions. Maybe sound could help out here. I think (and this is untested) that practice will only help so far. Certainly, we'd like to use real-world skills directly. Certain learned modifications of behavior (distance judgement, for example) seem to be hard to switch between in differnt contexts (i.e. VR vs. real-world). It has been shown that motor correction slows or stops with increasing motor to vision delays (up to 300 mS). Total systems delays of under 100 mS seem to be the goal to aim for. > >If you have references for these phenomena, I for one am interested. > I'll try to dig some up in the next week or so. My main reference list (in the back of a report I did awhile back) is out on loan right now. -------------------------------------------------------------------------- | My life is Hardware, | | | my destiny is Software, | Dave Stampe | | my CPU is Wetware... | | | Anybody got a SDB I can borrow? | dstamp@watserv1.uwaterloo.ca | __________________________________________________________________________