Google Maps uses DeepMind AI tools to help users predict estimated arrival times.

Google Maps is one of the company’s most widely used products, and its ability to predict future traffic jams makes it an indispensable tool for many drivers,media reported. Google says more than 1 billion kilometers of roads are driven by users every day with the help of the app. But as the search giant explained in a blog post today, thanks to DeepMind’s machine learning tools, its capabilities will become more precise.


In this blog post, researchers at Google and DeepMind explain how they get data from a variety of sources and input it into machine learning models to predict traffic flow. This data includes real-time traffic information collected anonymously from Android devices, historical traffic data, local government speed limits and construction sites, and factors such as the quality, size and direction of any given road. As a result, Google estimates that roads are better paved than unpaved roads, and algorithms think that sometimes it’s faster to take a longer stretch of highway than on multiple winding streets.

Google Maps uses DeepMind AI tools to help users predict estimated arrival times.

All this information is entered into neural networks designed by DeepMind that pick patterns from the data and use them to predict future traffic. Google says its new model has improved the accuracy of Google Maps’ real-time estimated arrival time by 50 percent in some cities. It also noted that with the outbreak of the new crown outbreak and subsequent changes in road use, it had to change the data used to make these predictions.

“When the blockade began in early 2020, we saw a 50 percent reduction in global traffic,” wrote Johann Lau, product manager at Google Maps. In response to this sudden change, we recently updated our model to make it more agile — automatically prioritize historical traffic patterns over the past two to four weeks and exclude traffic patterns from any previous time. “

The model works by dividing the map into what Google calls “super-roads” — adjacent street groups that share traffic. Each is paired with a separate neural network to predict traffic in the field. It’s not clear how big these supercars are, but Google says their size range is “dynamic,” meaning they change with traffic, and everyone with that feature uses TB-level data. The key to this process is the use of a particular type of neural network, the Graphic Neural Network, which Google says is particularly well suited for processing such mapping data.