Rethinking GPS: Engineering Next-Gen Location at Uber | Hacker News Reader

Rethinking GPS: Engineering Next-Gen Location at Uber

eng.uber.com

85 pointsmpweiher18 comments

Previous discussion: https://news.ycombinator.com/item?id=16887276

A big problem with Uber/Lyft is that drivers take all these fancy calculated positions and plug them into a standard navigation app, which can direct the driver to the completely wrong side of a large block. (Addresses are not very granular in a city.) It’s interesting to me why Uber/Lyft allow this.

The drivers also blindly follow this app even when the GPS position of the passenger has changed. I used to try and move to the right side of the street to make a pick up easier once I found out which direction the driver was coming from, but I don't bother anymore because the driver is following the nav app exclusively.

There's also the fact that Google Map data is just flat out wrong in many places in the middle of the country, it says my house is my neighbor's house and so on down the block, everyone is off by one.

I have tried to contact Google and get this fixed at least a dozen times over the past decade, always ignored. They used to have a whole community mapping thing where you could submit issues with the map, but those apparently are ignored.

It's a real nightmare trying to convince the insurance company and local government that google is wrong, love how that's my responsibility too.

It's funny, because out of the things that people complain about in modern navigation apps, "showing the right user location" is usually not that high on the list. There are so many other more pressing problems to solve. Map data is often inaccurate, bordering on unusable in less populated areas, and map data providers are colossally slow to update and correct their data. Navigation app UI is still surprisingly bad, even though this has been an area of fierce competition for at least a decade. Geocoding and reverse geocoding is still not perfect, but it's dependent on good map data. Accurate traffic is hard. Routing is pretty much solved but again, dependent on good map and traffic data, so it's perceived as needing improvement. Business search is hard because the underlying data changes so often, and for most nav system vendors, is not their core business so they ship something crappy. Most of these things are not improved by getting slightly better user position, so it's surprising Uber is investing so much into this research.

There's a UK company https://what3words.com/ that's been working on this for a while

> The use of particle filters necessitates a level of server affinity. Each new request to our service must be routed to the same back-end server for processing in order to update the correct particle filter.

Was implementing this on the client device considered? What were the motivations for this design?

My guess is a combination of the following:

1. Building 3D data is too large or inconvenient to store on the device.

2. Continuously downloading/caching small localized chunks of that data would result in unacceptable use of bandwidth for customers with limited data plans.

3. Their implementation of the algorithm may be CPU inefficient and contribute to excessive battery drain.

4. (cynical) This gives them yet another excuse to be sucking up and storing 1Hz precise user location data, potentially to use for other reasons...

Why do they need a 3d map to begin with?

Following their example of which side of the canyon the phone is, given the snr of sat b,c,d they can guess the side of the street.

If it's possible then being shadowed by an actual building, a construction site or a delivery truck amounts to the same result (but I guess the snr would vary between these 3 situations). Hence, why the need for the 3d model?

Long term thinking?

https://www.wired.com/story/uber-unveils-flying-taxi/

The example is simplistic, it's 3d space. It's not just which side of the street but buildings could be in any direction at various angles.

Any update on those new super-accurate GPS chips? [1]

1. https://news.ycombinator.com/item?id=15308771 Super-Accurate GPS Chips Coming to Smartphones in 2018 (ieee.org)

Amusing that they don't mention that they're able to better predict your location because of your phone's Wi-Fi.

Tracking people is necessary [0] as Uber needs to provide better services.

[0] https://www.npr.org/sections/alltechconsidered/2016/12/01/50...

Yeah, it's not like Uber did not hire Waleed Kadous away from Google, where he had led Android and indoor positioning, which make plenty of use of fused location. I bet he was involved in recruiting these folks (he approached someone I know in a related area. It seemed like they were investing heavily in every technology to sense the world, which makes sense for them).

Any competent phone manufacturer is already using Wi-Fi (and other input signals) to assist GPS, and providing the improved output to apps. So, if Uber is collecting Wi-Fi data, I doubt it's for the purpose of improving location.

Is this why Uber frequently pins me as being pretty far off from my actual position, even when something like Google maps is showing the correct position?

I mostly use Lyft now because I was tired of fighting with the Uber app to get it to show my correct location

I wonder why we don't have GPS transmitters installed at the top corners of every tall building. Seems to me like this would be a pretty straightforward solution in urban areas.

tl;dr using 3D maps to build a model of GPS shadows for receiver and satellite locations, and using the relative signal strength for each satellite, plus that model, to come up with a more accurate receiver location.

Original UCSB site: https://tia.ucsb.edu/about-tia/success-stories/shadowmaps/

A more detailed writeup by the original authors: http://insidegnss.com/auto/sepoct15-MADHOW_0.pdf

Reply on news.ycombinator.com