Creating Hyper-Accurate Maps from Open-Source Maps and Real-Time Data

https://eng.lyft.com/how-lyft-creates-hyper-accurate-maps-from-open-source-maps-and-real-time-data-8dcf9abdd46a

by maxerickson • 7 years ago

113 31 7 years ago

eng.lyft.com

While I may not be a fan of Facebook, Uber, and Lyft, I absolutely am a fan of their efforts to contribute back to OSM. Credit where credit due.

nexuist• 7 years ago

Lyft I would say probably fares the best out of that bunch on moral issues; driver treatment notwithstanding. That being said all of these companies have made amazing contributions not just to OSM but to the open source community in general - React has probably brought more people into software development than any existing web framework in history.

cookie_monsta• 7 years ago

Lyft I would say probably fares the best out of that bunch

I see what you did there.

kgwxd3• 7 years ago

They have a common enemy dominating the space. They have no choice, otherwise, they'd have taken it.

ryanjshaw• 7 years ago

I wish Uber had a similar model. Their maps think there is a road to my house that does not exist and they waste my time and their partners time sending people down a dead end that takes 5-10min to recover from. Reporting it on the app, the site, etc. goes into a blackhole and nothing gets fixed. I've even been told they use Google Maps, which they don't because Google Maps does not have this bad data point.

jsjohnst2• 7 years ago

I've even been told they use Google Maps

I could be entirely wrong, but it was my understanding Google Maps data is only used when a passenger is in the car, not when the driver is being dispatched to you (due to a licensing issue).

rathel• 7 years ago

I wouldn't be so keen. The local OSM mapper communities often find these corporations hard to work with. Lots of half-assed work, non-transparency in communication, doing armchair mapping (without local knowledge).

On the German OSM sub-forum there's a thread about improving OSM with data from Amazon drivers. It's hilarious how incompetent their mappers are and how sloppy their drivers are (e.g. reporting wrong speed limits).

rmc2• 7 years ago

Some of them had to be dragged kicking and screaming to do the right thing. Facebook initally dumped a pile of computer vision generated data into OSM, breaking everyone, and didn't tell anyone. It was entirely reverted ( more here: https://news.ycombinator.com/item?id=17856687 ). It took a lot of effort before they'd listen.

toomuchtodo2• 7 years ago

To get better at something, you must first suck at it. I believe the OSM oversight and governance model can support accepting contributions (data that would otherwise cost time and money from others to collect and contribute) from technology companies, while still ensuring guardrails exist to prevent poor data quality from being contributed (as exhibiting by Facebook and others working with OSM to ensure data quality contributed is high and it is done in a controlled manner).

alephnan• 7 years ago

What's the metric to be considered "hyper" accurate?

pj_mukh• 7 years ago

Random free idea: Grab image frames from the drivers' Phone camera and use that to figure out if the road exists where OSM doesn't have one. Supplement your own Kalman Filter and update your maps!

ivankirigin• 7 years ago

https://medium.com/@bluevisionlabs/blue-vision-labs-is-joini...

aloknnikhil• 7 years ago

The way iOS and Android implement image capture, you cannot have an app snap images without showing the viewfinder on the screen. This basically means the driver cannot do anything else except let the app take pictures while driving. Definitely not feasible.

kgwxd3• 7 years ago

Our first smart phone. Dad just bought it today. Do you have a smart phone?

aeleos• 7 years ago

Given the way that apps like Snapchat do an overlay of the camera image, in theory you could build an entire app over the camera. However it probably isn’t the best idea

cryptoz• 7 years ago

Has this changed on Android in recent years? I remember building a Camera app with a smaller-than-fullscreen viewfinder. I believe I got to set the dimensions of the viewfinder myself, making it as arbitrarily small as I wanted (or even optional?). Perhaps I remember incorrectly. I wonder if things changed.

trevyn• 7 years ago

This is definitely false from a technical perspective for iOS, though there may be something in the app review guidelines.

mariushn• 7 years ago

No need for cameras. Use existing vehicle tracking, via driver app GPS. Some phones have poor GPS, so wait until at least 2 vehicles pass the same route.

The issue isn't about missing roads. Most problematic are missing places when searching by business name or wrong/missing address numbers. (I work in the industry) Pelias project helps a lot, but there's still more data + software work to be done.

jsjohnst2• 7 years ago

Some phones have poor GPS

This is quite the understatement. I used to work for a company selling a personal safety device that paired with your phone and the data we would get back for location on some Android phones was just atrocious. On a large number of older Pixel devices, we’d routinely see location off by a large margin (as in over a mile in the worst cases). Interestingly, we’d see the “real” user location provided with a location accuracy of 100-400ft, then the device would radically jump to a new location (again, sometimes over a mile away) with a location accuracy of <50ft. When you are dispatching help to a person in an emergency, exact location matters!

Jommi2• 7 years ago

I saw a great hackathon project (which ended up winning 20K€) based on this, it was quite an interesting project at the time!

ganitarashid2• 7 years ago

If for nothing else but the sake of their open source contributions, I hope Lyft can find a working business model

kgwxd3• 7 years ago

Does Lyft not have the same business model as all other ride-sharing companies?

yodon• 7 years ago

That's the problem - they do

ganitarashid2• 7 years ago

Selling $5 bills for $1 is, unfortunately, not a business model

exogeny• 7 years ago

Cool! Now tell me how you'll become a profitable company.

Creating Hyper-Accurate Maps from Open-Source Maps and Real-Time Data

Related Stories