[yodon]

Correct. ARKit can keep the chart fixed above the car if you put it there, but it can't yet identify it as a car nor can it distinguish your car from your partner's car, and if you close the app and leave the lot and come back, you're going to need to place the chart over your car again in almost any realistic scenario. The best you get at present is "here is the floor" and "there is a wall", neither of which help the app provide contextual relevance for you. This doesn't make for terribly compelling charting applications beyond the first 15 seconds of "that's so cool it totally works" (which is a pretty cool 15 seconds).

The HoloLens will likely be able to keep the chart over your car if you leave your garage and come back, but not if you move your car and probably not if you park your car outside in the sun (which swamps the spatial projectors and prevents proper environment scanning), and you still need to have manually placed it over your car in the first place.

[lwansbrough]

I would suspect CoreML would play an important role in scene identification, which could certainly be used in conjunction with ARKit, don't you think?

Microsoft is investing a lot in computer vision, I'll be interested to see what changes they make to the next HoloLens' vision capabilities. Plane finding seems almost as good in ARKit as it was on HoloLens when I was developing on it (albeit the environmental understanding is limited/non-existent in ARKit, which as you mentioned is a big part of the equation.)

[yodon]

Fortunately plane finding is pretty easy even with something as simple as RANSAC[0]. I'm sure there are better algorithms now but ransac has been around forever and is nearly trivial to implement. It just doesn't easily generalize to "Siri, find my car"

[0]https://en.wikipedia.org/wiki/Random_sample_consensus

[alaskamiller]

Have an email? Would like to connect regarding ARKit