How feasible to build this HMD?

[UncleJoe]

http://research.engineering.wustl.edu/~ ... gahara.pdf" onclick="window.open(this.href);return false;

I recently read this and was thrilled by the prospect of near 180 FOV. The use of ellipsoid & hyperboloid mirrors allows both large FOV and a simple design.
The main downside I read was small exit pupil which darkens the image when you look at the periphery or any other eye movement.

Anyone knows how practical this design is and how feasible it is to build for DIYers?

subsystems:

1. LCDs + computer interface (use FPGA to decode DVI)
2. lenses
2. paraboloid & hyperboloid mirrors - I have no idea where to get these

[Okta]

That would be very cool but looks like a ton of work and expense. I guess a simplified version could yield decent results.

[cybereality]

Yeah, its a great concept. This came up before and there was a video of the project. Looked very cool.

I mean, it seems like a promising design but might be hard to build. Where do you even find mirrors like that and at what price?

I sometimes wish to build my own ultimate HMD but with this concept I wouldn't know where to start.

[UncleJoe]

I don't think it's reasonable for someone to build themselves, but a joint effort might be possible.


One thing that's worrying is how practical is this design? The report said small exit pupil was a problem, but are there any other show stoppers?
If it was practical, wouldn't some company have already marketed it?

One way to simulate such a design is to use a ray tracer. I've developed for PBRT before and I can create geometry to simulate the mirrors and lenses (PBRT actually has support for hyperboloid & parabaloid shapes - I thought it was just a curiosity, but wow, it does have use).

If someone knows or is willing to find out how to get such mirrors, I'd be willing to team up. I've used FPGAs before, so I could look into interfacing LCDs with DVI, but I don't know much about optics.

[cybereality]

Well I am not sure I can be of help but you certainly came to the right place. Lot of DIYers around here.

[crim3]

To me it seems a version of the leepvr design but with mirrors instead of lenses.
I mean, those spheric projections shown in the document reminds me the photo samples in leepvr site.
I have only done a quick look anyway.

[PalmerTech]

The construction is very feasible. The LCDs and lenses and control circuitry are no problem at all, and the mirrors could be custom built. There are several companies that make telescope equipment that can apple a front surface mirror layer to nearly anything, making the base mirror bowls out of fiberglass or aluminum or plastic would not be all that hard.

Here are the problems:

Look at the FOV chart again. They are relying on combing the FOV of the left and right eye, there is barely ANY overlap. Not only can that be uncomfortable and hard to use with normal drivers, it also means you only have 3D in that very small zone, right in the center. See how much effort they have to put into generating an image, just a static IMAGE, that would work with their HMD? No commercial DirectX drivers (Mostly IZ3D, really) support anything even close to that. Maybe practical for tele-presence applications that are custom built, but not useful for games.

Plus, that "wide" FOV is wasted when you cannot really look to the periphery of the field. This is also the main problem with a design like a laser based retinal projection system, it works fine without all these complicated optics, but you have to keep your eye pointing straight forward at the emitter.


Basically, this is neat, and very possible to build, but somewhere near useless for us.

[Johnny-Mnemonic]

This HMD have 60 degree horisontal overlap, where you will perceive stereo, this is actually more than diagonal FOV of any low-middle cost HMD's available on market, so this is not "only sweet-point" solution. You have quite a big area where you will be able to move eyes and see stereo.

What goes to drivers - yes, they need serious upgrade to be able to render proper images for right and left eyes. Any modern driver rely on 100% overlap, which is useless when we talk of high FOV. Anyway you are free to create your own software demo's which will benefit from this kind of HMD with proper stereo, fish-eye imagery etc.

[crim3]

What di you think if the LCD+lense is replaced with a pico-projector (that should be brighter than the LCD), and the ellipsoid mirror replaced with an ellipsoid white screen? That could avoid the "vignetting" problem, but at the expense of an aditional lense to work as eyepiece, I think...
(like in other occasions, just throwing ideas to get corrected by knowledgeable people)

To some extend, this design also reminds me of this:

You can find it here: http://www.mikesflightdeck.com/mirror_c ... splays.htm" onclick="window.open(this.href);return false;


I don't know if there are members with academic formation in optics.
I think that the rest of the knowledge areas needed to build a HMD are already covered at this forum.
I didn't believe that a DIY HMD could really be built, but seeing the efforts that are being done by some members and the high technical knowledge found at the forum the last year or so, now I think that it could be possible. Optics is the area that is left by now.

[UncleJoe]

Hurray. I've created a ray traced simulation of the HMD. Watch hmd_above_law.mp4 & hmd_express.mp4 I just bought some Steven Segal DVDs and what better way to enjoy the bone crunching glory! The video was kind of dark, so I made another version tuned to Madonna, Express Yourself.


To simulate the HMD, I created an ellipsoid & hyperboloid mirror in PBRT, and place the hyperboloid such that its virtual focus is at the same position as the upper ellipsoid focus. The viewer is positioned at the lower ellipsoid focus and looking ~45 degrees up. Horizontal field of view is 90 degrees.

Also, here are some pictures of the Matlab simulation I used to design the PBRT version:

planar_rays_only.png
only rays in Y-Z plane

thick_rays.png
shows the ellipsoid converging the diverging rays from the eye

angle.png
viewed from an angle


Look in hmd_simulation.zip for the code. Note that the stretched image near the bottom in the video is consistent with the higher ray density in the Matlab visualization. The next step would be to compute the projection on the film plane and find a transform to compensate.

Also, the small exit pupil problem needs to be investigated. I only rotate the eye in the video, but no translation.

[UncleJoe]

Apparently, there were 2 problems with the videos. Incorrect aspect ratio (1 instead of 1.6) and QuickTime incompatibility, which I've fixed.

Also, I noticed the transition between the orbiting motion and the final position, is very chaotic, which could be a wrong lighting model or inherently due to the geometry.

[UncleJoe]

It looks like the view will be very sensitive to eye movement indeed (see moving_eye.mp4). I've simulated moving the eye to the right by 0.1, and the reflected rays move a lot as was suggested by the chaotic behavior in the previous video. I'm not sure if small exit pupil is the right term since there's no aperture, unless you consider the monitor as the aperture.

The large movement is amplified by the relatively large distance between the hyperboloid and the monitor. The intersection of the rays and the hyperboloid don't seem to move much, so maybe I could try putting the film plane there like in the picture crim3 mentioned?

[3dpmaster]

It's impossible to make a perfect hmd like this. Nagahara has already stopped this ellipsoidal mirror project. The focal point changes constantly because of the rolling eyes.
This is a new project from the same university in Japan:
[youtube]http://www.youtube.com/watch?v=2rvDYF7XCjw[/youtube]

[UncleJoe]

Note: I only used a pin-hole camera when making the video, meaning only 1 ray from a point on an object will reach the film plane. The eye is definitely not a pin hole - the pupil is ~3 mm wide. I could redo the simulation to simulate a realistic camera with say a cone of 64 rays per pixel, which should reduce the small exit pupil problem, but probably with no practical significance.

[3dpmaster]

I think it would be perfect for augmented reality with half silvered mirrors.

http://research.engineering.wustl.edu/~ ... gahara.pdf" onclick="window.open(this.href);return false;