As a rideshare platform, Lyft’s core capability is to connect riders with drivers, just in time. This is a quickly evolving network where accurate estimates of ETAs, driving routes, and road dynamics are paramount for efficient operations. To deepen our capabilities in these areas Lyft set out to develop its own in-house turn-by-turn (TBT) navigation platform in 2019. Today, Lyft navigation is the primary TBT experience in the product and all mapping experiences are driven by the Lyft Map, an Open Street Map (OSM) derivative.
Building a viable navigation product is no small feat and our thesis from the start was that a) OSM’s road network today is sufficiently mature to support TBT navigation and b) that by repeatedly traveling on the road, Lyft could help improve and maintain OSM data efficiently. This bet succeeded and Lyft has become one of the top corporate OSM editors, making large contributions to missing roads, turn restrictions, gates and gate access, and highway destination and ramp signage. Lyft contributed more than 330k changesets in just the last two years, with 45k missing roads in 2023 alone.
To support our program, we built out an imagery collection platform that continuously captures refreshed road imagery, new methods that combine the road network, gps telemetry, and imagery to identify map errors, a batchless curation task allocation platform designed to continuously process detected map changes, and advanced tools for working with such large imagery collections in the context of mapmaking.
In this talk, I will review some of the technologies mentioned above that support an ongoing, cost efficient mapmaking operation and discuss insights and challenges.
