Google Maps is one of the company’s most widely-used products, and its ability to predict upcoming traffic jams makes it indispensable for many drivers. Each day, says Google, more than 1 billion kilometers of road are driven with the app’s help. But, as the search giant explains in a blog post today, its features have got more accurate thanks to machine learning tools from DeepMind, the London-based AI lab owned by Google’s parent company Alphabet.
In the blog post, Google and DeepMind researchers explain how they take data from various sources and feed it into machine learning models to predict traffic flows. This data includes live traffic information collected anonymously from Android devices, historical traffic data, information like speed limits and construction sites from local governments, and also factors like the quality, size, and direction of any given road. So, in Google’s estimates, paved roads beat unpaved ones, while the algorithm will decide it’s sometimes faster to take a longer stretch of motorway than navigate multiple winding streets.
All this information is fed into neural networks designed by DeepMind that pick out patterns in the data and use them to predict future traffic. Google says its new models have improved the accuracy of Google Maps’ real-time ETAs by up to 50 percent in some cities. It also notes that it’s had to change the data it uses to make these predictions following the outbreak of COVID-19 and the subsequent change in road usage.
“We saw up to a 50 percent decrease in worldwide traffic when lockdowns started in early 2020.”
“We saw up to a 50 percent decrease in worldwide traffic when lockdowns started in early 2020,” writes Google Maps product manager Johann Lau. “To account for this sudden change, we’ve recently updated our models to become more agile — automatically prioritizing historical traffic patterns from the last two to four weeks, and deprioritizing patterns from any time before that.”
The models work by dividing maps into what Google calls “supersegments” — clusters of adjacent streets that share traffic volume. Each of these is paired with an individual neural network that makes traffic predictions for that sector. It isn’t clear how large these supersegments are, but Googles notes they have “dynamic sizes,” suggesting they change as the traffic does, and that each one draws on “terabytes” of data. The key to this process is the use of a special type of neural network known as Graph Neural Network, which Google says is particularly well-suited to processing this sort of mapping data.