Hacker News new | comments | show | ask | jobs | submit login
Magic Leap One Video – Diffractive Waveguides Confirmed (kguttag.com)
174 points by IntronExon 8 months ago | hide | past | web | favorite | 85 comments

Very interesting insight from a former GPU developer at Magic Leap:


He believes they should have focused on making the headset and gone with integrating with a mobile device instead of an integrated SoC...That said there's usually a reason (low-level) engineers are engineers and not product people... but he makes some good points.

The one I most disagree with is "why develop Apps with ML when you could develop apps on Apple (iOS)"

Well one good reason is that a) game developers and VR/AR devs aren't necessarily developing on iOS already yet and b) it's an entirely new (potentially large) new market where you could become a top developer more easily. And they'd still have to build a significant part of their SDK on top of iOS or Android, no?

It's also a very niche market and you end up completely captive in their ecosystem if you do this. It's reasonable to assume that there'll still be a market for iPhone and Android apps 5 years from now, I don't even know if Magic Leap One will still exist in 5 months or if they'll ever achieve commercial success. Yeah it'll probably be vastly easier to be the top ML developer but it won't do you much good if you end up with 12 users and a dead platform.

IOs/Android development is reasonably mature now and they're backed by huge companies (and other smaller companies who make dedicated tools for that ecosystem). It would make sense to build on top of that instead of starting from scratch.

Also, price-wise you'll never be able to compete with smartphones when it comes to bang-for-the-buck. The economies of scale are huge. These AR/VR need as much processing power as you can throw at them, it's very hard for a small team to come up with a competitive offering, especially for an embedded wearable solution. If they were making their own custom SoC with specialized hardware to enhance the AR experience it might make some sense, but as far as I understand they're only using off-the-shelf components. It does seem like a strange design choice.

> It's also a very niche market and you end up completely captive in their ecosystem if you do this.

During my masters in Interaction Design I kind of "accidentally specialised" in gesture interfaces (the pointing kind, not the touchscreen kind). So obviously I have an old-school Kinect, and later got a Leap Motion[0]. This is exactly the problem with the latter.

Your only real options are either doing research, or doing some kind of custom job like designing an AR/VR installation for a museum.

[0] https://www.leapmotion.com/

You can use the Leap Motion in AltspaceVR and Bigscreen already, I hope the Tabletop Simulator developers will support it natively with intuitive gestures one day.

> The one I most disagree with is "why develop Apps with ML when you could develop apps on Apple (iOS)"

I could see that making sense if it means you can ship earlier. At this point, it's starting to feel like the AR/VR hype machine has flipped from breathless optimism to cynical skepticism. If ML ends up shipping some $2000-$3000 dorky looking headset people will laugh and the product will die.

A big barrier to VR is the fact that it doesn’t work out of the box and needs configuration. My VR headset is collecting dust since I don’t have time to reconfigure it. If you make a device that works as fast as a Nintendo switch (put it on head, turn on and it opens the game) you will have a Really compelling device .

Eh? The setup process on Rift is a 5 minute operation, granted the Vive takes about twice as long. Don't see this as significant either way though, not many people drop $400-800 on a piece of kit and then don't bother to use it because they're too lazy to configure it.

You're ignoring having to mess around with your PC every time you want to use the Rift/Vive.

What do you mean by "mess around"?

Well, I don't own a Rift, but I assume you can't just walk up to your PC while it's in sleep mode or whatever and without touching it, put a Rift on your head and automatically continue playing a game you had been last playing two days ago (Something that they've promised the "Oculus Go" will be capable of)

If your PC is on and your Rift connected you don't need to touch anything it detects when you put it on and throws you right into the menu we're talking about 5 seconds from it detecting you're putting it on to hitting the menu.

Honestly have massive respect for the Oculus team and their user experience, if I had to boot up Steam every time I used VR it would start to fall into more of a chore.

I definitely see value in stand alone devices for these reasons and for the casual users, but I do also see value in having top end PC power in a tethered device for creatives and serious gaming.

I think you can if you use steam big picture. EDIT: actually what you are describing is exactly what Big Picture solves.

Exactly, the thought of having to move the sensors around for optimal performance (I can't leave the 2 rear sensors with cables on the floor out in the office), and recalibrate each time translates to the Oculus not being used.

This problem is already going away with "inside out tracking". Many of the Windows Mixed Reality headsets already do this, no external sensors to calibrate as with the Vive/Rift. Oculus have already demonstrated prototypes with inside out tracking. I think VR has many other problems, but this one will largely go away very soon.

It sounds like you might really like the announced (but not released) Oculus Go. Standalone device, no PC required, inside out tracking, somewhere between a GearVR and a PC connected Rift.


Presumably Magic Leap is developing hardware and software that will be highly optimized for AR/computer vision and will go far beyond the capabilities of an iPhone or Android device. If ML was to come out and say tomorrow they are relying on the iPhone to run their headsets and apps, their valuation would tank. It's built on the premise that ML has some type of secret sauce that will be much better and more immersive than what's out there on the market right now.

If they were going to use a 3rd party dev ecosystem and build their SDK on top of an AR dev tool, I'd expect Unity would be the top choice. Apple didn't really have an AR-focused dev tool until they publicly released ARKit in the middle of last year, making that somewhat of a non-starter for a company that has been around as long as ML.

Would it be really feasible to use either Bluetooth or WIFI as the connection between glasses and the CPU/GPU? Sounds like quite challenging setup if you want to provide a smooth user experience. There's a lot of data coming from the sensors on the headset and graphics data going back and much of this is pretty time critical.

With iOS, you will likely also run into issues with background processing. On mobile phones, the battery life is always an issue if you plan to do CPU/GPU intensive stuff.

Sounds much better approach to make the glasses smart and capable of working to some extent independently. Then you can have a higher level protocol between the glasses and phone.

I would imagine the latency at the required (high) resolutions would make WiFi / BT unusable.

There's an AR kit for IOS which would make development super easy if you already know the basics behind IOS app development and are a game developer (which I imagine is a sizable population).

There seems to be several people with a tech hardware background that doubt that Magic Leap can deliver on their promises. But somehow they have no problem raising money or getting publicity. Either they have something that is real and extremely convincing to investors that they have been able to keep very secret, or they have a really slick slide deck made of lies.

We’ll know for sure if/when the first production devices go on sale, but if Shaq was wearing a real pair then I don’t want them.

Theranos had no trouble getting funded and even Walgreens to trust them. They were eventually found out to be charlatans. Just because you can raise money doesn't mean you have anything to back it up. When people are skeptical they sometimes have a grain of truth to it. From what I have read about Magic Leap they seem to have vaporware. Even if they didn't the VR / AR space is getting more and more competitive so the longer they don't release even if they have something it may not even matter.

Usually when people present something that is new an innovative you can understand how they did it. Take for example the new iPhone camera with Face ID. They gave enough details for someone to understand generally how it was done. You can look at the physical pieces of it to see how they did it. Sure Apple may have made some claims that were bending the truth to market it but the technology has applications other than security (animojis for instance). Also, it was an extension of previous technology.

When people do the same analysis with Magic Leap they find that they are outright lying and that they can't actually have anything to backup their claims. Instead they are trying to pass off technology that doesn't have the advantages that they state as something completely different.

Walgreens (and CVS) both also sell homeopathic stuff of all kinds, all over.

Even charlatans can sell stuff and make a ton of money. "Can raise money" or "can trick people into buying" (remarkably similar, really) can not in any way be construed as a validation of a product.

Agreed. Just getting on a store shelf doesn't mean you have a good product, it just means that store believes customers will pay for that product.

Like you said, Walgreens, a major pharmacy specializing in real medicine, sells homeopathic crap. They also sell very unhealthy food and alcohol. They also sell cigarettes. Not because they believe the product is good, but because that product makes them money.

The problem is there are huge gaps between having a great idea, creating something really impressive, and bringing a product to market successfully and selling at scale. That last step is the most difficult of all. Very few companies pull it off.

The best I can tell, once a dev kit is released they still have 3 or 4 years before they will be selling their hardware at a meaningful scale - if everything goes alright. Their burn rate is way too high, guessing by how many people they hired. They will need to have an IPO or they are expecting to raise Uber level funds. Maybe that’s realistic but I wouldn’t bet on it.

>They will need to have an IPO or they are expecting to raise Uber level funds.

neither is possible with no revenue or only dev-kit revenue. you are overlooking the third option: do an ICO.

That might be the most realistic solution, ironically. They might end up in the bin of start up death from pre-mature scaling.

I have no doubt that someone is going to win big in AR. Apple can hit the ground running with all of these ARKit apps. Unless Apple really screws something up, they will have the software advantage by a wide margin.

I'm not convinced that either Google's Series B investment in 2014 or Sundar Pichai's board seat indicate that Google believe's Magic Leap has a high probability of success. The series B was now quite a long time ago. Google has had no trouble getting in to the same line of business as their own investments or companies were the CEO sat on the board.

Or get purchased by a larger entity

If you look at Ponzi schemes or con artistry, a little skepticism and a little peer pressure help part people from their money.

In fact, if you're at all familiar with con, the best marks are the people who think they'll never be a mark and have a little bit of greed in them.

And a key part in this is that a lot of experts openly claimed that Theranos was selling snake oil. You can't do those kinds of tests with that little blood.

High expected value can make investment with low probability of success a good investment. Hard to implement and develop optical VR technology could launch company that has similar dominance as Intel has had in the CPU markets.

If the underlying development path of Magic Leap is promising, investors may keep Magic Leap going for a decade even if they don't deliver anything.

Investors include Google, Qualcomm, Alibaba, Kleiner Perkins and a16z.

Huge expected value can also make people fools, without a scrap of real hope of future returns. A lot of money has been flushed down a lot of glittery drains by many a smart person/org in the search for wealth. When you add subtle yet incredibly challenging technical hurdles buried deep within the hype, the odds of flushing greatly exceed fortune.

It’s not as though Google, to pick one example from your list, is some kind of techno-prophet; they’ve killed more projects than asteroids did dinosaurs.

> Huge expected value can also make people fools, without a scrap of real hope of future returns.

See also, Gambler's Ruin.


That's only an issue if you are making bets such that a large loss will make you go bust. If you are facing that possibility, then you need to move down a level. If you can continue to make bets at your level, then the +EV will increase your roll over time.

Poker lessons.

You don't have to know if any particular company has real changes or not, what you need is is reasonably good estimate of how likely you are a fool.

That estimate made in ignorance of the relevant technology is all the answer you need as to foolishness.

> Huge expected value can also make people fools, without a scrap of real hope of future returns

The probability of failure should already be factored into the term "expected return"

I think Clinkle and Juicero have shown us that viability and product are only tangentially related to being able to raise VC money

> Either they have something that is real and extremely convincing to investors that they have been able to keep very secret

Hate to be that guy, but wasn't Clinkle the last top-secret-many-vc-much-money startup that went bust before the product was even released?

Look how long Theranos was able to keep it up despite experts calling b.s. from basically day one.

Yes, but that's why healthcare investors stayed away from Theranos - that should have been a sign that the core tech wasn't workable.

But MagicLeap had huge funding rounds led by Google and Alibaba. They presumably did their due diligence and involved people who know what they're doing. That suggests to me that MagicLeap is doing _something_ that's real. I'm gonna wait and see.

I suspect that Google Ventures isn't made up of engineers.

I'm not sure what difference it makes but it was actually Google, not Google Ventures, that made the investment.

> Google is placing a big bet on it: in addition to the funding, Android and Chrome leader Sundar Pichai will join Magic Leap's board, as will Google's corporate development vice-president Don Harrison. The funding is also coming directly from Google itself — not from an investment arm like Google Ventures — all suggesting this is a strategic move to align the two companies and eventually partner when the tech is more mature down the road.


It is ironic to see that Sundar Pichay has degree in metallurgy and material science. I expected something better from him.

For me, it is hard to believe to that a man with effectively a physics degree bought into that.

Enron is a prime example for keeping things secret for long time.

Bernie Madoff is another standout example, and the many scams in the supposedly expert world or wine, such as Rusy Kurniawan.


I watched a documentary about this guy (though I can't remember the name). What a totally fascinating story! The fact that even today there are people who think that he wasn't involved in forging wines really speaks to how easily and thoroughly people can get caught up in a convincing lie proliferated by a convincing liar.

Sour Grapes I believe.

Or the Moller Skycar.

People mention Theranos, I'd also add uBeam - probably the longest running joke in the Silicon Valley.

Apparently investing doesn't require belief in laws of physics.

These guys were definitely going to be using diffractive waveguides (photonic crystal waveguides). Look at the LinkedIn profiles of some of the folks on their optical design team (especially the ones with titles that include "Diffractive Optics Engineer"). One had a PhD adviser at Berkeley (Eli Yablanovitch) who originally coined the term "photonic crystal", another wrote a widely used open source implementation of RCWA (S4) while working with a top group in photonic crystal design at Stanford (Shanhui Fan), and some of their engineering leadership came to the company from what was Digital Optics Corporation.

It might not have been clear to investors who were fed patents with fiber scanning all over them.


I Think its also quite telling that they found the biggest person they could find (Shaquille) to do the physical showcase. On an average sized person these will be massive, and not just look(and feel) like large weird sunglasses. And no, I dont believe that these where the "large" version, except in the regard of the headband. I doubt the core unit will be meaningfully smaller.

Where in comparison, the HoloLens team seemed to take more care in the design of a large device on a person's head. Looking strange but functional.

You also have to consider that they were interviewing the commissioner of the NBA and Shaq is still one of the biggest personalities associated with the league. He does a ton of on air promotion for the sport.

It's a big headset, but I don't believe the set out to find the biggest person just so they could deceive people to think they aren't big.

I'm still waiting for more info on Near Eye Light Field devices.[1][2]

Actually, it appears there's more on this now[3][4], but the video's audio is a weird translation of some paper or something.

1: https://research.nvidia.com/publication/near-eye-light-field...

2: https://www.youtube.com/watch?v=uwCwtBxZM7g

3: http://research.nvidia.com/publication/2017-11_%09Near-eye-L...

4: https://www.youtube.com/watch?v=NdTycenXID8

> 2: https://www.youtube.com/watch?v=uwCwtBxZM7g

> "In 1968, Ivan Sutherland introduced the first graphics driven headmounted display, of HMD"

Ivan Sutherland? As in: the guy who made Sketchpad? What hasn't this guy done?

That was painful to read and the comments are even worse. I'm not sure I trust anything stated in this write-up.

I love this guys blog, he does fantastic tear downs of display technology and I just don't see anyone else doing it in the space. He called out the 85% shade when they presented the initial preview a couple months ago.

He is genuinely not impressed by much and maybe he should be in some cases, but if you look at his body of work he's not just some yahoo brigading on magic leap hate and if there's a better technical voice on display tech out there, I haven't seen it.

While true, there's just no way you can infer the opacity from those frame grabs.

On the one hand he says he says he used the whites of Shaq's eyes as a reference and on the other he states that the design of the glasses hugs the users head to block out ambient light getting in from the side.

There's a slew of issues with this derivation. To arrive at a number like 15%, without showing his homework, is as questionable as the marketing claims of Leap themselves.

I'm not saying he's wrong but it does sound like a lot of bias.

He is saying that it is an approximation (see in comments): "I looked at what I thought were comparable regions. I checked several frames of the video and the frame at 0:51 in the article is what I used. The method is admittedly very approximate as you have to guess the “gamma curve” of the video capture device to back out the linear value (I used a gamma of 2.2)."

Why? His grammar could use a bit of work, but his write ups are often in depth and quite accurate. He’s hardly an unknown in the world of optics, and his previous analysis of Magic Leap have all proven accurate.

For example



I think some of the issues pointed out with his assessment in the comment section hold merit. His "no your wrong" response without any real evidence as to why doesn't sit well with me. That said, if he's well known for his knowledge perhaps that's enough. I personally have no idea who he is.

As with his proofreading, I’m not terribly interested in his comment section people skills. His decades in the field, his track record of tear downs and predictions, and his expertise do interest me. I’m noting a paucity of technical, rather than personal counterpoints here, or in those afformentioned comments on that level. Loads of appeals to authority, and a kind of, “well they must have something valuable!” credulity abounds.

He certainly can't simply be dismissed. OTOH he was recently seen promoting the idea that Magic Leap can't get camera-based inside-out positional tracking working, which is probably quite preposterous.

I hadn’t seen that, can you link me to it?

He doesn’t seem to make any claim about what they can’t do regarding tracking, he just says that there are likely IR markers in the demo. The counter argument as usual seems to come down to awe at how much money ML raised and from whom, as well as the talent pool they bring.

None of this is to say they have perfected it, or anywhere close. That will be a long way off. We have struggled with ambient IR light and glass since the days of the Wii and Xbox Kinect, and we still struggle with it now, and will probably continue to struggle with it for quite some time. It may well rely on getting LIDAR or something along those lines in the headset to truly provide reliable tracking in all environments.

That from the op of the Reddit thread, after explaining his incredulity thst ML wouldn’t have tracking down. Meanwhile Karl doesn’t seem to appeal to anything other than released material, patents, and physics.

I’m inclined to beleieve him.

> He doesn’t seem to make any claim about what they can’t do regarding tracking, he just says that there are likely IR markers in the demo.

From Guttag's article:

> Adding markers to the frames is a crutch to make up for poor registration and SLAM (simultaneous localization and mapping).

> The counter argument as usual seems to come down to awe at how much money ML raised and from whom, as well as the talent pool they bring.

Hang on: earlier you seemed to be expressing incredulity that Guttag had claimed ML didn't have acceptably-working SLAM by now. Is your position now that in fact Guttag was correct to make this claim?

IR markers would seem to hint at reliability issues of tracking. Presumably one could reply "algorithms are still being developed and fine tuned".

But if it's not reliable, then I can't see how it would be usable by the mass market.

People don't generally respond well to half of their visual world desyncing from their physical body, even if it's 1:1000 seconds.

Getting this whole "silicon valley tv show" vibe here, where it's the VR company that gets to much hype. Almost seems like they are too deep in and need to keep going hoping it will be accepted in the market.

Still think AR has a lot more potential than VR in the long run. Just need to get that "Yes!" moment where the technology has a perfect fit and use for it.

I’d argue that VR has more potential. Once we get to OASIS level VR, it’ll kick the ass of having a HUD in your glasses.

Let's hope we won't have to physically move around, and it's slightly more like Altered Carbon VR :)

Side-note: The Vive pro has two cameras.

"""The rest of the video made my ears bleed. It was the usual Rony doing his hype and buzzword salad trying to fit in as much hype in as he could without really saying anything."""

Hype and buzzword salad, I'm going to be taking that one with me into the future.

No need you bother to take it with you, I'm sure there is going to be plenty.

It seems like the way they made things appear to have substance in AR then is to drop down visual light to 15%.

Maybe they reproject the view onto the screen in order to have (effectively) passthrough VR while still looking like they have AR?

Pretty doubtful. One of their original claims was to produce blacks, which is really hard in AR as you need to selectively block light. It initially sounds easy until you realize your light blocking mechanism needs to be in focus. If you use a lens to put it in focus, the things around it are now out of focus, etc.

If you can't produce blacks, things behind AR objects bleed though. Your objects appear as sci-fi holograms instead of real objects. Reducing pass through light by 85% reduces bleed through by 85% and makes your objects appear 85% more real.

What does "more VR than AR" even mean? Either the user is seeing virtual objects overlaid on the real world or not. It has nothing to do with the opacity of the lenses.

I've found the Meta 2 AR headset to be quite dark (and it's still not quite as dark as what was shown in the pictures in the article) and it becomes very difficult to look at the graphics and understand them as part of the real scene behind them. With the real surroundings so darkened out, it creates a completely different contextual space. You definitely get to a point that you forget to pay attention to the real space and only pay attention to the visualization on its own.

So that's what "more VR than AR" means. Continuity of context between real and virtual is the defining attribute of AR. Without it, you just don't have AR.

If you can't see much of the real world (15% light penetration by this account), how useful is that overlay? If it was 1% light penetration it would look almost completely black, to the point you may not detect any real world at all. At 99% light penetration, you probably can't even tell there's any shading at all. Somewhere in the middle different people will draw the line as to what's usable as AR and what isn't (and even then, it may depend on the action being attempted).

Yeah but it's not 1%. At 15% your eyes will adjust. Why bother with 15% if the point is not to overlay on your surroundings?

> Why bother with…

Because they might have no choice. The million (billion?) dollar question is: are they full of bull? That’s what this is all about.

He’s saying: they want to go higher than 15%, but they can’t. Not ‘they choose not to.’ They would have if they could. And they can’t go 0% and full VR because that would be admitting defeat, and the end of the gravy train.

I'm pretty sure this is deliberate. Remember when they said they could make projections opaque? And they said some waffle like even though it is impossible in theory, in practice you can do it.

It looks like their solution is to make the glasses dark, so that in comparison the display is very bright. If it is bright enough compared to the background you won't see what is behind it. Making the glasses dark is probably ok because your eyes will adjust (your eyes have an insane dynamic range).

Also, realistically even with AR you spend most of your time looking at the augmented part rather than reality (based on my experience with the hololens anyway).

AR Glasses are going to be as big as the Apple Watch.....

So pretty good then? Apple Watch is doing extremely well.

Would Apple still even bother to sell the Watch if they had the Apple ideal version of AR glasses.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact