How Uber Knows Your Driver Is 7 Minutes Away

Uber's driver location tracking system relies on the idea of road segments rather than entire roads. The platform maps around 100 million road segments, each with a time estimate for crossing. This time estimate is not static; it varies throughout the day based on traffic conditions. Rather than purchasing traffic data, Uber utilizes the existing location pings from its drivers every four seconds as a free source of traffic telemetry.

Each time a driver crosses a segment, Uber records the precise time it took, thereby creating a live traffic sensor network without the need for additional cameras. A deep learning system called DeepETA was introduced in 2022 to predict ETAs for the next three hours, taking into account both historical data and live conditions. DeepETA can adapt to real-time events like accidents or different traffic patterns on quieter streets by comparing the live data against historical patterns.

The routing engine then sums up the segment crossing times to calculate the ETA, which goes through an additional layer of correction based on real trip data to ensure accuracy. This sophisticated system has proven successful, with shipping DeepETA improving long-trip arrival accuracy by 6%, which is estimated to contribute around $100 million in annual gross bookings for Uber.