304 North Cardinal St.
Dorchester Center, MA 02124
304 North Cardinal St.
Dorchester Center, MA 02124
For this article I would like to dive down on the most likely display and optics Magic Leap (ML) is developing for their their Product Equivalent (PEQ). The PEQ was discussed in the “The Information” story “The Reality Behind Magic Leap.” As I explained in my November 20, 2016 article Separating Magic and Reality (before the Dec 8th “The Information” story) the ML patent application US 2016/0327789 best fits the available evidence and if anything the “The Information” article reinforce that conclusion. Recapping the evidence:
I can’t say with certainty that the optical path is that of application 2016/0327789. It is just the only optical path in the ML patent applications that fits all the available evidence and and has a chance of working.
Rony Abovitz, ML CEO, is claiming a larger a larger FOV. I would think ML would not want to be have lower angular resolution than Hololens. Keeping the same 1.7 arc minutes per pixel angular resolution as Hololens and ODG’s Horizon, this would give a horizontal FOV of about 54.4 degrees.
Note, there are rumors that Hololens is going to be moving to a 1080p device next year so ML may still not have an advantage by the time they actually have a product. There is a chance that ML will just use a 720p device, at least at first, and accept lower angular resolution of say 2.5 or greater to get into the 54+ FOV range. Supporting a larger FOV is not small trick with waveguides and is one thing that ML might have over Hololoens; but then again Hololens is not standing still.
The support of vergence/accommodation appears to be a paramount issue with ML. Light fields are woefully impractical for any reasonable resolution, so ML in their patent application and some of their demo videos show the concept of “focus planes.” But for every focus plane an image has to be generated and displayed.
The cost of having more than one display per eye including the optics to combine the multiple displays would be both very costly and physically large. So the only rational way ML could support focus planes is to use a single display device and sequentially display the focus planes. But as I will outline below, using sequential focus planes to address vergence/accommodation, comes at the cost of hurting other visual comfort issues.
Both high resolution LCOS and DLP displays use “field sequential color” where they have a single set of mirrors that display a single color plane at a time. To get the colors to fuse together in the eye they repeat the same colors multiple times per frame of an image. Where I have serious problems with ML using Himax LCOS is that instead of repeating colors to reduce the color breakup, they will be instead be showing different images to support Sequential Focus Planes. Even if they have just two focus planes as suggested in “The Information,” it means they will reduce the rate repeating of colors to help them fuse in the eye is cut in half.
The Hololens which also uses a field sequential color LCOS one can already detect breakup. Cutting the color update rate by 2 or more will make this problem significantly worse.
Another interesting factor is that field sequential color breakup tends to be more noticeable by people’s peripheral vision which is more motion/change sensitive. This means the problem will tend to get worse as the FOV increases.
I have worked many years with field sequential display devices, specifically LCOS. Based on this experience I expect that the human vision system will do a poor job of “fusing” the colors at such slow color field update rates and I would expect people will see a lot of field sequential color breakup particularly when objects move.
In short, I expect a lot of color breakup to be noticeable if ML support focus planes with a field sequential color device (LCOS or DLP).
An important factor in human comfort is the latency/lag between any head movement and the display reacting can cause user discomfort. A web search will turn up thousands of references about this problem.
To support focus planes ML must use a display fast enough to support at least 120 frame per second. But to support just two focus planes it will take them 1/60th of a second to sequentially display both focus planes. Thus they have increase the total latency/lag from the time they sense movement until the display is updated by ~8.333 milliseconds and this is on top of any other processing latency. So really focus planes is trading off one discomfort issue, vergence/accommodation, for another, latency/lag.
Another issue which concerns me is how well sequential focus planes are doing to fuse in the eye. With fast movement the eye/brain visual system is takes its own asynchronous “snapshots” and tries to assemble the information and line it up. But as with field sequential color, it can put together time sequential information wrong, particularly if some objects in the image move and others don’t. The result will be double images, getting double images with sequential focus planes would be unavoidable with fast movement either in the virtual world or when a person moves their eyes. These problems will be compounded by color field sequential breakup.
I don’t know all the behind the scenes issues with what ML told investors and maybe ML has been hemmed in by their own words and demos to investors. But as an engineer with most of my 37 years in the industry working with image generation and display, it looks to me that focus planes causes bigger problems than it solves.
What gets me is that they should have figured out that focus planes were hopeless in the first few months (much less if someone that knew what they were doing was there). Maybe they were ego driven and/or they built to much around the impression they made with their “Beast” demo system (big system using DLPs). Then maybe they hand waved away the problems sequential focus planes cause thinking they could fix them somehow or hoped that people won’t notice the problems. It would certainly not be the first time that a company committed to a direction and then felt that is had gone to far to change course. Then there is always the hope that “dumb consumers” won’t see the problems (in this case I think they will).
It is clear to me that like Fiber Scan Displays (FSD), focus planes are a dead end, period, full-stop. Vergence/accommodation is a real issue but only for objects that get reasonably close to the users. I think a much more rational way to address the issue is to use sensors to track the eyes/pupils and adjust the image accordingly as the eye’s focus changes relatively slowly it should be possible to keep up. In short, move the problem from the physical display and optics domain (that will remain costly and problematical), to the sensor and processing domain (that will more rapidly come down in cost).
If I’m at Hololens, ODG, or any other company working on an AR/MR systems and accept that vergence/accommodation is a problem needs to be to solve, I’m going to solve it with eye/pupil sensing and processing, not by screwing up everything else by doing it with optics and displays. ML’s competitors have had enough warning to already be well into developing solutions if they weren’t prior to ML making such a big deal about the already well known issue.
The question I’m left is if and when did Magic Leap figured this out and were they too committed by ego or what they told investors to focus planes to change at that point? I have not found evidence so far in their patent applications that they tried to changed course, but these patent applications will be about 18 months or more behind what they decided to do. But if they don’t use focus planes, they would have to admit that they are much closer to Hololens and other competitors than they would like the market to think.