Google doesn't use Android phone location data as effectively as they could for privacy reasons. Tracking people's location in aggregate to make decisions about which roads are usable at which speeds is okay, but as soon as you get to the levels of "only one guy went along this road in 2020, should we mark it as private", it becomes a privacy issue.
It's a privacy issue to use that kind of data, because the employees who build the map shouldn't be looking at one guys GPS traces. So instead the traces are cut into tiny chunks and aggregated and only shown where there are more than 50 people at the same spot. That effectively means there is no data available to make decisions on the most rarely used roads.
First, I’m not sure how much they would be concerned about privacy issues of only a single person over a year. They uniquely identify individuals over multiple sessions. I think they also wouldn’t need to know the specific identity (“John Doe”) and could just use their anonymized id (a unique individual whose name we don’t know).
Also, even if they choose to not do this for maps even though they do it for lots of other things, merely aggregate traffic info on roads would be useful. 364 days of zero traffic and one day with 30 minutes of one car would be useful.
> So instead the traces are cut into tiny chunks and aggregated and only shown where there are more than 50 people at the same spot.
The same algo that does that could tell the field mapper to go look at roads that are barely used (or whatever criteria ends up being the best), it doesn't need to tell them how many people or who went on that road.
Sometimes I think that Google create their road maps solely using satellite pictures.
Plenty of people with their Android phones will still go over that section with their snowmobiles, four-wheelers, bike, etc.
How does Google tell the difference between someone on an electric bike vs a car going on a dirt road doing 20-30km/h to not have rocks flying everywhere?
I ran into this problem a lot when I was living in downtown SF of all places. One should never drive in front of AT&T/Oracle stadium when an event is ending, as the traffic is insane and part of the road is closed to allow for the mass of pedestrians. But Maps interprets pedestrian traffic as car traffic and shows that section of the road as green and moving faster than the sections with actual cars, and tries to direct you down it.
Obviously we need to have Google connect our cars to the internet so they have more data points. And we also need Google to connect our bikes to the internet, that way Google can tell the different between a bike and a car. And we need to prevent the bike from being able to be used if the Google tech isn't charged and working, because it would be dangerous for everyone else if Google thought that person was in a car and not a bike.
You jest, but in time this will be true, not because of Google, but insurance requirements. Same for autonomous cars. You would still be able to drive your car yourself, if you can afford the insurance, and you won't.
I couldn't find a paper that talks about the specifics of this very problem. The ones I've found in the past fail with outliers like a bike going through stopped traffic and or the scenario above.
The best approach I could think of is they expand the time they're looking at to classify the vehicles before the outlier situations.
I believe you can get a lot of information from accelerometer data. Especially when compared with other data for the same road. Is it even used in this way?
