Skip to main content
Uber logo

Start ordering with Uber Eats

Order nowOrder now

Start ordering with Uber Eats

Install the appInstall the app

Enhancing the Quality of Uber’s Maps with Metrics Computation

July 12, 2018 / Global
Featured image for Enhancing the Quality of Uber’s Maps with Metrics Computation
Figure 1: The Uber Map Model is a data structure made up of features and attributes.
 Requirements of Uber’s map quality computation system:
Figure 2: On this map, GPS traces that match roads are shown in blue, while GPS traces that do not match a road are shown in red.
Figure 3: In this example, our map suggests a route, but drivers, responding to turn restrictions, must take a different route.
Figure 4: Refining access points is composed of three main steps: identify actual pick-up and drop-off locations used by drivers for a place or address (left), compute the distances from those locations to the place or address (middle), and then set the preferred access point based on the shortest distance (right).
Figure 5: We define geographical areas as S2 cells and use those cells to build data sets which we can process for map quality.
Figure 6: Using map data and geographical regions as inputs, our metrics computation system partitions the data onto S2 cells, then uses those cells to compute its metrics.
Figure 7: As roads and other map features change, our metrics computation system works to continuously keep our metrics up-to-date, ultimately delivering a better user experience on the rider and driver apps.
Ines Viskic

Ines Viskic

Ines Viskic is a senior software developer on the Maps team at Uber and has, along with her colleagues on Uber’s MapQuality team, developed the metrics computation framework outlined in this article. She holds a PhD in Computer Engineering from University of California, Irvine.

Posted by Ines Viskic