Magic Leap is a Tragic Heap


The title of this review was carefully chosen, not glibly.  I want what is best for VR and all other technologies on the Reality–Virtuality Continuum, Magic Leap included.  Unfortunately, their current offering is a tragedy in the classical sense, even more so when you consider how their massive funding and carefully crafted hype sucked all the air out of the room in the AR space.  It is less of a functional developer kit and more of a flashy hype vehicle that almost nobody can actually use in a meaningful way, and many of their design decisions seem to be driven by that reality.  It does not deliver on almost any of the promises that allowed them to monopolize funding in the AR investment community.

There are lots of general overviews of ML1 at this point, so I will be focusing on a few specific points that have not been as widely covered.  If you want to check out a broad overview, this video from Tested is a great place to start.  If you want to see the guts and how they work, I got to help iFixit to tear down my ML1.

The Controller

Image result for magic leap controller


Tracking is bad.  There is no other way to put it.  The controller is slow to respond, drifts all over the place, and becomes essentially unusable near large steel objects – fine if you want to use it in a house made of sticks, bad if you want to work in any kind of industrial environment.  Magnetic tracking is hard to pull off in the best of cases, but this is probably the worst implementation I have seen released to the public.  For VR enthusiasts who are familiar with systems from Polhemus (the OG magnetic tracker people), the Razer Hydra, or the ever-elusive Sixense STEM, you know where the bar is.  From the Magic Leap developer manual: “6DoF tracking is stable in slow to moderate motion. It also recovers and relocates quickly in case of fast or sudden motion (e.g. boxing or fishing – like movements).”

I understand that Magic Leap wanted a controller that did not require line-of-sight to the headset or a bulbous protrusion to report position, but that was a terrible tradeoff to make, especially for developers who need a controller that just works – there are good reasons no other company decided to walk this path.  Parlor tricks like holding the controller behind your back are fun, but ML1 could and should have used basically any other type of tracking system.  Several other companies have managed to pull off inside-out optical tracking without Magic Leap’s billions of dollars in funding, and if they can’t pull that off, they could definitely use a system with an external reference to get things going.  As things stand, a lot of their software and UI limitations seem to be driven by the bad controller.

In another bizarre departure from competing devices, the trackpad is not clickable.  The Steam Controller, HTC Vive wands, Oculus Go, Lenovo Mirage Solo, etc all have a clickable trackpad, and designers have heavily relied on that feature.  Even the Playstation 4 controller has one!  What this means in practice is that selection via the touchpad requires either lifting and tapping (terrible for precision) or clicking the trigger while holding (also terrible for precision).  It also means that they cannot use the trackpad to emulate buttons or other selection schema. Everyone else in the industry is using components from ALPS (great company BTW), they should have just called them up and told them they needed a custom trackpad with fun RGB LEDs.

Final note on controllers: Unlike most magnetic tracking systems, the transmitter side is in the controller.  This means there is a giant ferrous core wound with copper wire hanging out just above the trigger.  To balance things out, Magic Leap had to install metal counterbalance weights in the bottom of the controller.  That does make the controller feel “premium” at first heft, but really sucks for long term ergonomics.

The Computer

Image result for magic leap lightpack


They call it the “Lightpack”.  It is basically the guts of a tablet computer in an oversized hockey puck that you wear on your belt.  This is the best part of the device by far, A+!  I would have expected Magic Leap to do the fashionable thing and throw all their render hardware and battery power on the headset itself for looks, but some group of sane people appear to have recognized that putting your heaviest components on the most weight sensitive part of your body is a bad idea if you want people to actually wear your product for any period of time – this is a longer topic for another day, but the data shows that you need to be BRUTAL when it comes to reducing HMD weight.  This approach also allows them to use much more powerful chips than they could feasibly cram into a head worn device.

The cables are robust, and the weight tugging at the back of your head actually helps counterbalance things a little.  They should have made the battery replaceable, but nobody is going to use their ML1 long enough for that to matter to anyone but collectors with an aim to preserve the history of AR and VR.

The Headset

Image result for Magic Leap


They call it the “Lightwear”. This is the part that has gotten the most hype over the years, with endless talk of “Photonic Lightfield Chips”, “Fiber Scanning Laser Displays”, “projecting a digital light field into the user’s eye”, and the holy-grail promise of solving vergence-accommodation conflict, an issue that has plagued HMDs for decades – in other words, ensuring that the focus of your eyes always matches their convergence, something that Magic Leap has touted as critical in avoiding “permanent neurologic deficits” and brain damage.  It is even more important for AR than VR, since you have to blend digital elements with real-world elements that are consistently correct.  

TL;DR: The supposed “Photonic Lightfield Chips” are just waveguides paired with reflective sequential-color LCOS displays and LED illumination, the same technology everyone else has been using for years, including Microsoft in their last-gen HoloLens.  The ML1 is a not a “lightfield projector” or display by any broadly accepted definition, and as a Bi-Focal Display, only solves vergence-accommodation conflict in contrived demos that put all UI and environmental elements at one of two focus planes.  Mismatch occurs at all other depths.  In much the same way, a broken clock displays the correct time twice a day.

In more detail: The ML1 uses six waveguides all stacked atop one another, three for each RGB color channel at two different focal planes.  You can think of it as a Bi-Focal Display – a display system that can move the focus of the display between two different values based on eye-tracking with no variability in between, unlike continuously vari-focal displays such as Oculus Half-Dome or Nvidia’s true lightfield displays.  I don’t have exact measurements yet, but it seems like the near-plane is focused at about 0.75 meters, and the far-plane is focused at about 5 meters.  If they stick with this technology (and I have not seen any indication that they can do anything else, especially not their much-hyped fiber display), each additional focus plane will require even more stacks of waveguides and infeasibly high framerates (each plane consumes a minimum of 60hz from the temporal budget).  I don’t think that will happen within reasonable weight, image quality, and cost constraints. 

More than one plane is good, don’t get me wrong!  It allows developers to avoid extreme mismatch with very close or very far objects.  That said, spinning hype and monopolizing investment with promises that cannot be met is bad for the entire XR industry, not just Magic Leap.  Hardware manufacturers have a responsibility to clearly communicate the capabilities of their hardware to developers, even when the capabilities fall short of what they would prefer.

On to the other parts of the headset: The tracking is good compared to most other players in the AR/VR industry, but worse than most of the big guys, including Hololens.  Expect jitter in ideal environments.  If you want a comparison, think halfway between PSVR and Rift.  The meshing system is good, but not nearly as fast as Hololens.  It is pretty similar to what you see from companies with a few orders of magnitude less funding, like Stereolabs.

Bi-focal abilities aside, the image quality is acceptable.  Have you seen Hololens?  Think that, but with slightly larger FOV.  The rainbow artifacts are a bit worse owing to the large number of stacked waveguides and the black levels are a bit better, but Magic Leap is playing in the ballpark as everyone else.  Despite drawing enough power to keep the headset nice and toasty (seriously, it is hard to touch the magnesium shell if you are in a warm room), the display is far too dim to use outdoors.  That is a shame, since the transparency is about the same as a pair of dark sunglasses – not exactly indoor material.  How does the eye-tracking work?  Impossible to say, because nothing uses it.  That is not a great indicator.

A true leap would have been a FOV that is wide enough to be useful, something that Magic Leap could have done if they prioritized user experience over meeting device size expectations.  For a good example, check out Dreamworld at 90 degrees – the tracking is not at all comparable, but the experience is pretty exciting.

The Operating System

Image result for magic leap OS


Magic Leap says they have “built a whole new operating system” called LuminOS to take advantage of their “spatial computing system“.  It is actually just Android with custom stuff on top, the same approach most people take when they want to claim they have built a whole operating system.

I will keep this part short.  I hope Magic Leap does cool stuff in the future, but the current UI is basically an Android Wear watch menu that floats in front of you.  The menus are made of flat panels that can only be interacted with through the previously discussed non-clickable trackpack.  Eye tracking and rotation/position of the controller are ignored, as is headlook.  You can toss Windows 8 style application windows all over the place, floating in space or even attached to walls!  That is nifty, mostly useless, and also exactly what Microsoft started showing off about three years ago.  It is some of the worst parts of phone UI slammed into some of the most gimmicky parts of VR UI, and I hope developers create better stuff in the near future.

Sales Figures and Developer Adoption

Image result for dozens of us


The Magic Leap order system was really easy to figure out for the first few days after launch.  I gathered some order numbers from friends and compared their order times, and I am pretty confident about predicting first-week sales.  Unfortunately, they changed the system shortly after I tweeted about it.  Based on what I do know, it looks like they sold about 2,000 units in the first week, with a very heavy bias towards the first 48 hours.  If I had to guess, I would put total sales at well under 3,000 units at this point.  This is unfortunate for obvious reasons – I know over a hundred people with an ML1, and almost none of them are AR developers.  Most are tech executives, “influencers”, or early adopters who work in the industry but have no plans to actually build AR apps.  This was a big problem in the early VR industry, and that was with many tens of thousands of developers among hundreds of thousands of development kits sold!  Multiplying the problem by a couple orders of magnitude is going to be rough for ML.


Conclusion

Image result for magic leap wired


Magic Leap needed to really blow people away to justify the last few years.  The product they put out is reasonably solid, but is nowhere close to what they had hyped up, and has several flaws that prevent it from becoming a broadly useful tool for development of AR applications.  That is not good for the XR industry.  It is slightly better than Hololens in some ways, slightly worse in others, and generally a small step past what was state of the art three years ago – this is more Hololens 1.1 than Consumer AR 1.0.  Consumer AR can’t happen without advancement, and it seems those advancements will be coming from other companies.  There is, of course, a chance that Magic Leap is sandbagging us; maybe the real deal is just behind the next curtain!  Past experience suggests otherwise…

Above is a telling picture from a piece Magic Leap did with Wired magazine a couple years ago, back when they were still hyping up scanning fiber displays.  See the fancy-looking, high-tech light up  strands?  They don’t do anything.  It is just electro-luminescent wire.  It looks great to casual observers, but does not hold up to any kind of scrutiny from people who are in the know. If you want to try using it to dress up  your own costume, gaming PC, or multi-billion dollar hype machine, you can buy a nice assortment for $20 here
———————————————————————————————————–

Update: About 45 minutes after I posted this, the CEO of Magic Leap decided to share his love of the television series Avatar: The Last Airbender.  He also points out that the photographic stereoscope was invented in 1838, very interesting!