It's a hard problem, I agree -- how can the app tell the difference between "missed a turn; send the person around the block" and "driver rejected that decision"? There are some intersections, ramps, blocks, etc that I avoid, and I'd like to be able to tell the app "don't do that to me". Sometimes a route works fine at some times of day but not others; on one of my first days driving home from the new office the app kept trying to get me to make a turn that was temporally illegal -- one of those light-up signs. And it doesn't seem to know about the construction that has a key block near my workplace closed; on my second day it kept trying to bring me to the same intersection over and over via different routes, and I didn't know the area well enough to recognize what it was doing for a while. (You may recall that roads in Pittsburgh are not always straight and grid-like. :-) )
Here's something I've been wondering about: my phone already (in general, I mean) notices if I say "OK Google" -- so it obviously always has an ear open, so to speak. Now I don't want the navigator app listening to noises in the car unbidden; that would be creepy, even if it might learn things from the occasional "dammit", snarl, etc. However, could it respond to "OK Google, no left turn" or the like? Or even, more simply, "OK Google, mark" combined with a way for me to later review and annotate the places where I said "mark"? As a drive I want to provide "on the ground" feedback that improves navigation, and given that the software knew where I was when something interesting happened, it seems like we could connect those dots.
Granted, this also means needing to be able to process arbitrary input like this, some of which will be garbage or wrong. You're trying to do this at scale, so you're looking for machine learning, not human verification. Maybe you can ask other local people to confirm things, much like my phone asks me assorted questions about restaurants I've been geo-located in (questions like "is this a good place for vegetarians" and "is there a good view"). I'm not the only person who's driven around that closed road the app doesn't know about; maybe you can put that to use somehow?
Just some rambling thoughts. I know it's a hard problem.
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
(no subject)
Date: 2018-04-20 01:17 am (UTC)Here's something I've been wondering about: my phone already (in general, I mean) notices if I say "OK Google" -- so it obviously always has an ear open, so to speak. Now I don't want the navigator app listening to noises in the car unbidden; that would be creepy, even if it might learn things from the occasional "dammit", snarl, etc. However, could it respond to "OK Google, no left turn" or the like? Or even, more simply, "OK Google, mark" combined with a way for me to later review and annotate the places where I said "mark"? As a drive I want to provide "on the ground" feedback that improves navigation, and given that the software knew where I was when something interesting happened, it seems like we could connect those dots.
Granted, this also means needing to be able to process arbitrary input like this, some of which will be garbage or wrong. You're trying to do this at scale, so you're looking for machine learning, not human verification. Maybe you can ask other local people to confirm things, much like my phone asks me assorted questions about restaurants I've been geo-located in (questions like "is this a good place for vegetarians" and "is there a good view"). I'm not the only person who's driven around that closed road the app doesn't know about; maybe you can put that to use somehow?
Just some rambling thoughts. I know it's a hard problem.