Improving Accuracy of Maps Using Anonymized GPS Traces
eng.uber.comS2 Geometry [1] is a pretty cool library for anyone who hasn't need it before and it doing map stuff. They mention it in the article half way through. Allows you to programmatically bucket GPS data and quickly search adjoining cells, etc. I was using this for a wifi project a while back and was pretty awesome.
GPS traces aren't anonymous. If you go the same places often, your GPS trace is easily traced to you.
I don’t know about Uber, but Apple has spoken about collecting similar data and they anonymize it by slicing the route into a bunch of little pieces, throwing out the beginning and end of the route, and anonymously uploading the bits in the middle.
See “Probe data and privacy” section here: https://techcrunch.com/2018/06/29/apple-is-rebuilding-maps-f...
There's no indication that they are submitting the full trip trace. If they only submit 500 meter long traces, it may he sufficiently anonymous.
I might have missed it, but how do they handle roadwork in European cities?
They have a lot of one way street and an entire path might be reversed for 2 weeks during utility work, then brought back to normal (like replacing cast iron gas pipes, or moving all the utilities before a light rail).
I guess the idea here is that Uber is developing technologies that would put them ahead of a new entrant/competitor. Time will tell.
What map data does Uber use? I think they're making heavy use of Google Maps' API for the front end but what about the routing - do they roll their own or use Mapbox/Google?
"driver-partner" yuck.
I thought the same thing. Also was interesting in Fig 2, where the map was wrong, it’s all driver-partner...yet in Fig 3, where it was an illegal turn, they conveniently drop the -partners.
See also: “student-athlete”