I think ARKit-style SLAM will turn out to be a fad, they'll be a lot of interesting toy apps, but AR without environmental understanding or persistence I don't think offers much beyond that. The basic ARKit demos we've seen are the same stuff we've been seeing for a while now demoed with third party libraries.
Including depth sensing HW was the right solution, but Google doesn't have its own popular smartphone as a forcing function. I predict eventually Apple will include a depth camera, or they'll use dual-cameras to try and synthesize it, and once that happens, then all Android manufacturers will follow suite.
If AR is to be useful, it's got to be a lot better at tracking and drift, at making sense of the world, of supporting occlusion and mapping.
That's interesting. Regarding HW vs SW, I think the exact opposite.
I always thought Kinect Tango and other HW-dependent solutions were a fad, and SW-only on ubiquitous cheap sensors is the right solution. SW keeps getting cheaper, faster than HW does.
I predict that, even if Apple (re)starts the custom-HW fad, within 10 years we'll see devices with good-enough camera-plus-SW AR outnumber specalized-HW AR devices by at least 10:1, and it'll only get more extreme over time.
(Mind you, that prediction requires that we continue to get performance-per-watt-per-dollar improving; i.e. it assumes that Moore's law finds a way to outlive current processor fabrication technologies.)
> If AR is to be useful, it's got to be a lot better at tracking and drift, at making sense of the world, of supporting occlusion and mapping.
ARKit seems like the near term future. It's useful and no longer requires as much expertise since is available in the OS and is well documented. For a software library it's surprisingly accurate.
The Tango model of using extra hardware is probably much better, but seems further ahead. The software model works today on existing devices and lets people see how this is useful. Once you have that convincing people it's with the money to have the hardware added to the phone becomes easier.
Given how many Android phones are lower cost than the flagship was Tango ever going to be very popular? Apple could have forced the issue (like Lightning or the headphone jack) but people could always switch Android OEMs to get something cheaper if they don't see the value.
Tango's extra hardware was for depth sensing, and low-energy feature tracking, but the basic technique of plane detection from what I understand, is the same technique ARKit uses, which came from Flyby according to one article I read. In 2014 when Tango was released, even Apple HW wasn't powerful enough to run the tracking in software alone.
I have a feeling the end game of this is going to be that Tango-like devices are used for mapping the world, and ARKit like libraries consume the geometry.
That is, not everyone has to own a device with depth sensing. For example, if Streetview-like services using LIDAR, or if self driving cars with LIDAR, map point clouds of most outside areas, and merchants and vendors map into areas with specialized Tango-like HW, then most of the benefits of depth sensing AR can be had for people without depth sensors.
It would be a mostly static 3D map of the world, not frequently updated, but probably good enough to enable a large number of apps.
> but people could always switch Android OEMs to get something cheaper if they don't see the value.
The kind of people that buy flagship Android phones would probably either see the value or be price insensitive enough not to switch over it.
OTOH, “works ok now” is often more important than “works better later”, so getting something out that will work with today's flagships has value even if Tango would be practical down the road.
ARKit is a perfect example of "worse is better". Quite obviously inferior to the full Tango demos with occlusion and room mapping, and HoloLens, but simple enough to excite the imagination, and to enable "fake AR" experiences like Pokemon Go.
That's what I was trying to get at. Once people get a taste I think the demand for More capable solutions like Tango will be much higher than it would have been otherwise.
When Pokémon Go came out I was very disappointed to find it's much hyped 'AR' was really just rendering ok top of a live camera image with no tracking at all.
The demo of the ARKit version from WWDC is what I had been expecting.
I think the android market has proved that there are a lot of users who don't pay attention to that. Otherwise we wouldn't have so many comments and stories about the lack of updates or how long it takes vendor X.
I'm just thinking about your average person who goes into at Best Buy or cell phone store and wants a new android phone. Given to similar phones, one with the extra hardware and the other cheaper, if they don't see the point they'll probably go with the cheaper one.
If they got to use software only AR stuff on their previous phone that may change that decision.
Including depth sensing HW was the right solution, but Google doesn't have its own popular smartphone as a forcing function. I predict eventually Apple will include a depth camera, or they'll use dual-cameras to try and synthesize it, and once that happens, then all Android manufacturers will follow suite.
If AR is to be useful, it's got to be a lot better at tracking and drift, at making sense of the world, of supporting occlusion and mapping.