Usually the navigation app on your phone can help you understand the current road conditions, but you can’t predict what will happen in the next 10 minutes, 20 minutes, or even 50 minutes. In this link, artificial intelligence plays a role. Google recently announced on its official blog that Google Maps has launched DeepMind’s new machine learning model to predict future traffic congestion in real time.
Google said in a blog post that with the new model, Google Maps has improved its real-time forecast arrival time accuracy by 50 percent in some cities. “Our model is already very accurate in predicting time, remaining accurate on 97 percent of trips, and by working with DeepMind, we can further reduce the percentage of inaccurate forecast arrival times for Graph Neural Networks.” Google said in a blog post.
Google says Google Maps has improved its real-time forecast arrival time accuracy by 50% in some cities
The improvement of the predictive accuracy of this model is due to the comprehensiveness of the data. According to Google, the new model includes real-time traffic information collected anonymously from Android devices, historical traffic data, speed limits from local governments and construction sites, and information about the quality, size and direction of any given road.
DeepMind, an artificial intelligence star owned by Alphabet, Google’s parent company, has long wanted to combine its technology with practical applications. According to DeepMind, in this collaboration with Google Maps, the company designed a model that divides the map into “super sections”: adjacent streets that share traffic. Each super-section is paired with a neural network that can predict the flow of that segment. It’s not clear how big these supercars are, but Google says their sizes change “dynamically,” suggesting that they change with traffic and that the amount of information on each segment reaches trillions of bytes. The key to the whole process is to use Graph Neural Networks, which DeepMind says is particularly well suited to processing such mapping data.
Google says two data are critical throughout the model: authoritative data from local governments and real-time feedback from users. Authoritative data can let Google Maps know if speed limits, tolls or certain roads are restricted by buildings or new crown outbreaks. Real-time feedback from drivers can help Google Maps quickly show whether a road or lane is closed, whether there are buildings nearby, whether there are disabled vehicles or objects on the road, and so on. At the same time, the data can help maps understand unexpected changes in road conditions affected by mudslides, blizzards, or other natural forces.
With the help of this model, users can use the most efficient driving routes. For example, when a model predicts that AC traffic in one direction may increase, a bill of lading scheme with low traffic flow is automatically found for the user. The model also considers road quality factors, and if the road is not paved or covered with gravel, dirt, and dirt, the model also allows the user to choose an alternative.
Back in real life, you may not have to worry about road conditions in the future, but can go to the appointment as scheduled. When you leave home, traffic may be unobstructed and undisturbed. But Google Maps can predict and live traffic, and you can see that if you continue on your current road, you’re likely to get stuck in unexpected traffic jams within about 30 minutes, which means users are likely to miss appointments. So Google Maps will use its knowledge of nearby road conditions and events to automatically re-route users to help them avoid traffic jams and make appointments on time.