Google Maps helps users navigate over one billion kilometers in more than 200 countries and territories daily, and Google says its estimated time of arrival (ETA) predictions have been consistently accurate for over 97 percent of trips. That’s not good enough for Google, though, so the company integrated DeepMind AI tech for machine learning that can make its ETAs even more accurate.
Before partnering with DeepMind, an Alphabet AI research lab, Google Maps used a combination of historical traffic patterns and live traffic conditions to understand current traffic patterns. The partners wanted to be able to predict future traffic patterns, so DeepMind developed a graphic neural network, which also considers data on the time of year, road quality, speed limits, accidents and closures.
Thanks to that machine learning approach, Google Maps has improved the accuracy of real-time ETAs by up to 50 percent in places like Berlin, Jakarta, São Paulo, Sydney, Tokyo, and Washington D.C. Now, Google Maps can warn users about traffic jams before they exist.
As we saw with COVID-19, unprecedented events can drastically disrupt traffic patterns and throw off prediction models. Google saw up to a 50 percent decrease in worldwide traffic when lockdowns started early this year. The sudden change forced Google Maps to be more agile. It began prioritizing historical traffic patterns from the last two to four weeks and deprioritizing older patterns.
Google says predicting traffic and determining routes is incredibly complex, and it will continue looking for ways to keep users out of gridlock and on the safest, most efficient routes possible. Most recently, Google Maps has added in-depth spoken walking information, expanded its AR walking directions and added crowd predictions to bus and train routes.