The elephant in the room is that UWB positioning is indeed very precise, but only if you have a number of well-positioned beacons with known precise locations working together. The article is all giggly about the rosy future, but having worked with UWB (specifically, the Decawave chips mentioned in the article) I would be much more restrained with predictions. Yes, this is really good technology, but nowhere nearly as easy to use as it seems.
I'd be very interested in technical details: are they using 802.15.4a? Do they use the spectrum above 6GHz?
I know very little about any of this stuff, but it occurred to me that Apple already has a few technologies in play by which devices can sense the layout of their surroundings: HomePod can determine its proximity to walls by listening to how sound reverberates in a space, and iPhones map depth by comparing data between by dual cameras as well as by dot projection. Could these technologies potentially be used to eliminate or reduce the need for multiple UWB devices?
You don't need to know the positions of the beacons unless you want an instant location fix. You can use SLAM with inertial odometry or visual inertial odometry to find a precise position using UWB beacons at unknown positions. SLAM algorithms will also give you the positions of the beacons. The catches are that it requires compute, and it requires you to move around for the model to converge.
Is it any good for the where-is-the-dog use case in the article, where you might want to find something at opposite side of the house, presumably without a network of beacons to help?
Yes, to a certain extent. Frequencies above 6GHz are easily attenuated, especially by anything which contains water (so, human bodies), but bags and clothes are fine.
I'd be very interested in technical details: are they using 802.15.4a? Do they use the spectrum above 6GHz?