Navigation apps like Google Maps aren't just about keeping us from getting lost. They're also a handy tool for knowing the speed of the road and avoiding speed cameras. Another important feature is choosing the best route to avoid traffic jams or knowing what to expect in terms of duration. But how does Google Maps know about real-time traffic?
When you open the app, you enter your destination and automatically receive information about the different possible routes, the estimated travel time, traffic density and the expected arrival time. This is a lot of information that requires a lot of work and essential accuracy.
Thanks to information provided by users themselves
According to Google, real-time traffic data comes from users themselves. Every time you use the app, you provide information about your location. When this information is analyzed in bulk, Google can determine traffic conditions on the roads of the whole world.
By studying how your phone's location changes over time and whether that change is slower than usual, Google can determine if you're in a traffic jam. Your data will be combined with that of other users around you.
For example, if the normal speed on the road is 110 km/h, and Google detects that several phones on the road are moving at only 20 km/h, it deduces that there is a traffic jam.
A few years ago, artist Simon Wecker had fun walking around Berlin pulling a cart full of 99 smartphones and managed to trick Google into thinking there were traffic jams. The speed and density of phones fooled Google's traffic prediction algorithms. Google Maps.
A combination of historical information and machine learning
But this is not its only tool for accurately determining traffic in real time, because this information would not be enough to determine the evolution of road flows over time.
To do this, it uses analysis of historical traffic patterns over time. To be clear, Google knows that a particular road tends to have a certain flow of vehicles between certain hours. Take Paris exits, Google knows that late Friday afternoon traffic tends to be heavier for weekend departures combined with work exits.
By cross-referencing this information with live traffic conditions and machine learning (machine learning algorithm ), it is possible to generate projections on the evolution of the state of the roads.
Google worked with DeepMind, an AI company owned by Alphabet, to improve its real-time traffic forecasts. Using graph neural networks (GNNs), Google was able to calculate arrival times (ETAs) with over 97% accuracy. The collecting more historical measurements over time allows him to continue to refine his technique, although changing trends, as happened at the start of the COVID-19 period, have also posed a challenge to the reliability of the forecasts.
Google knows not only the flow of a road on a day-to-day basis, but also its appearance: whether it is paved or not, its width, its layout, the landmarks it passes through… this data is important when it comes to suggesting an alternative route. It is not only a question of knowing whether the road will take you from point A to point B, but also of know the quality that it offers. After all, it may be more efficient to wait a few extra minutes on a congested highway than to take a clear secondary road.
Data from third parties
Finally, Google also uses information provided by the relevant administrations to determine real-time traffic. This corresponds to official data on speed limits, tolls, limited access roads or roads whose lane is closed due to works.
As with Waze, Google also takes into account real-time user feedback, such as the presence of accidents, stopped vehicles or strange obstacles on the road. Administrative and user data can be used to qualify and provide insight into exceptional events on a road such as an accident or fog.
To provide an ever more reliable estimate of traffic in real time, Google Maps therefore relies on a certain number of data at its disposal: position and speed data provided by users via their smartphone, historical data optimized by the machine learning algorithm and official data from third parties.
Of course, the forecasts may never be 100% reliable, but one thing is for sure, they will continue to improve!
Article updated on January 10, 2025 by Byothe
