Can confirm CPU part. I once tried to replace a 2d stepwise linear function interpolations (basically, a loop finding the right step, then linear interpolation, on a double; the function's objective is to rescale inputs before they become regression's arguments) by LUTs.
All of this looping was happening in another tight loop, and there were multiple LUTs.
It was a failure with overall decreased performance and increased cache misses. Plus, I could never get the same precision.
I went through 2 PCVR headsets and I'd never think: oh this is going to be mainstream (in the present state!).
Those headsets are not comfortable (sweaty); the hardware needed for anything realistic is top tiers; it takes PhD/time to configure (and goes out of sync!); the software ecosystem was(is?) not super stable; the interaction (controllers, tracking) is sometimes awkward (controllers are big, interfere with keyboard and real world). Often it small things like comfortable physical head position vs. position in VR world are not "lined up"
We'd laugh that it takes as much time to get into the VR plane simulator as to drive to the airport to fly real plane!
Compare a VR kit to a monitor pair, or a game controller (imo VR is kind of a cross of the 2) - both are mostly plug an play.
I'm changing my simracing rig from VR to triple-monitors for this reason. (And also the narrow FOV). When it works it's amazing. At least I can pretend the sweaty visor is because I'm wearing a racing helmet. But every 2 weeks or so something messes up and I have to spend an hour or two resetting things and reinstalling things.
Also only a few simracing games have really good VR support.
I'd add a (major) inconvenience is problem investigation, often exceptions point you to root cause and all you need it to break on the first exception (think after running big codebase, for an 1h just to get where problem is), but if they are thrown willy-nilly this debugging strategy is MUCH less convenient...
Tokyo: In 2007, _65_percent_ of trips within a 50 mile radius were by mass transit. Overall transit usage is [...] approximately double that of all combined usage in the United States and nearly 10 times that of Paris [http://www.newgeography.com/content/002923-the-evolving-urba...]
It is a small team, maybe few people? On the other hand, Amazon has 10k people on Alexa[1]. One wonders how much is worth info they are sucking out of households....
That could be it, I suppose. Sorry, I'm not trying to make this a personal attack on the team (I don't know any of them), and I'm not personally invested in their success or failure (I haven't purchased any of their stuff or invested in them). I just think that it would be cool to have an easy-to-use, hackable home assistant that does most of its computation locally, and that's what they're (supposed to be) building.
In a lot of ways, it doesn't really matter why they're not producing results, so much as the fact that they're not producing them. I mean, check out the list of remaining action items in their last "Production Update"[1]. This came over 8 months since shipping what were supposed to be the Mark II devkits, and almost four years(!) after they announced work on the Mark II in early 2018 [2], which has since gone through about a zillion different design changes. I won't claim to be an expert, but having had even a bit of experience with product development of this kind, this is not the happy path to getting consumer electronics shipped.
I use MyCroft/PiCroft (Pi4) + PS2 mic array (15$), it is cheap and relatively easy to hack in Python/Git. I believe it uses Mozilla's voice service. Has no ads.
Commands which work nicely which we use all the time: "Hi Mycroft, set timer to X minutes." "play news", "set alarm to 6 am", "what's the temperature?", "will it rain?", "what's the time in Paris?".
I wrote METAR and TAF module for it to get more detailed weather (and learn some Python).
All of this looping was happening in another tight loop, and there were multiple LUTs.
It was a failure with overall decreased performance and increased cache misses. Plus, I could never get the same precision.