Machine learning to improve maps and navigation?
The average American motorist probably can’t recall maps the last time they purchased a paper map, printed out a digital map, or even asked someone else for directions. We no longer have to worry about getting lost because of the availability of GPS and mobile mapping apps on our smartphones, which provide us with up-to-the-minute direction suggestions.
However, this is not the case in regions of the world that are developing or experiencing rapid economic growth. Commercial mapping services like Google Maps, Apple Maps, Bing Maps, and the like simply can’t keep up with the rapid expansion of Doha, Qatar’s road network over the past five years.
The United States probably cannot understand the scale
“Each one of us who grew up in Europe or the United States probably cannot understand the scale at which these cities grow,” says Rade Stanojevic, a senior maps scientist at the Qatar Computing Research Institute (QCRI), part of Hamad Bin Khalifa University, a Qatar Foundation university in Doha. Every few months, it seems like a new underpass or overpass or major highway is built through some neighborhood.
According to Stanojevic, an expert in network economics, a road network can thought of as a huge graph in which each intersection represents a node and each road represents an edge. Characteristics of a road segment can either static, like the posted maps speed limit, or dynamic, like the amount of traffic during rush hour. A machine-learning model would only need a large amount of up-to-date data on both the static and dynamic factors to see where traffic is actually going, rather than where an old map says it should go, and then predict the best routes through an ever-changing maze. Today’s vehicle fleets, thanks to their monitoring systems, generate a wealth of data, as Stanojevic puts it.
In Doha known as Karwa
Stanojevic has discussing taxis. His group at QCRI collaborated with a taxi service in Doha known as Karwa to gather comprehensive GPS data on all of their maps trips. Using this information, they developed a new mapping service called QARTA, which provides routing advice to drivers at Karwa and other operators like delivery fleets.
According to Stanojevic, QARTA has a better grasp of the real-world road and traffic situation in Doha, allowing drivers to shave off seconds from each maps trip and increasing fleet efficiency by five to ten percent. According to Stanojevic, “if you’re managing a fleet of 3,000 cars, 5% of that is 150 cars.” “You could take 150 cars off the road and not notice a difference in sales at all.”
Although QCRI’s system is unlikely to able to compete with the major map-services providers in the developed world, Stanojevic says it may help cities in maps the Middle East and other developing regions better manage growth.
The big picture to reduce emissions
Today we’ll talk about how to improve mapping in rapidly expanding cities. Traffic. Congestion and an ever-increasing number of vehicles mean that even maps the best-intentioned navigation apps can’t make driving directions easier. But imagine if your country’s population doubles in just ten years from now. Now that there are new streets, communities, and structures, a new map required.
Dr. Rade Stanojevic, a senior scientist at the Qatar Computing Research Institute at Hamad Bin Khalifa University, a Qatar Foundation university, is here maps with me today. Dr. Stanojevic is an expert in the field of network economics and computer networks. To that end, he is currently attempting to construct more precise models of real-world traffic in Doha, Qatar, and other cities by employing graph theory, machine learning, and other techniques.
Spanish companies before joining QCRI’s team
However, the road system has some dynamic features. Things like traffic volume and average speeds vary by time of day and weekday. There are some occurrences maps that simply defy prediction, etc. What makes this problem interesting and useful to everyday life, and in particular the business cases that we deal with, is understanding both the underlying static nature of the road network and the dynamic parts that come from the traffic.
They are ineffective because they were not designed with the maps understanding that Doha’s infrastructure would evolve as rapidly as it has. At QCRI, we came to the conclusion that network science and machine learning hold the keys to answering many of these questions. As a result, a few of us began investigating automatic map inference.
The network’s inner workings
It was sometime in 2017 when we first began working on this, maps and we quickly learned that this is a very difficult but critically important issue for many emerging urban centers. Indeed, we made significant headway in this area, too, by uncovering the network’s inner workings. In addition, we later realized that the map also possesses dynamic properties that tied to traffic.
It took about 18 months for a major intersection that had changed sometime in 2016 when we moved, when I moved to Doha, only a maps few hundred meters from our office, to reflected in Google Maps. This meant that for roughly 18 months, that intersection was completely hidden from Google Maps. As a result, drivers would forced to take a lengthy detour that wasn’t necessary to avoid the intersection altogether if those alternate routes weren’t available.
The quality of Google Maps and similar services has also increased over time. They’ve identified the issue and have shortened the 18-month lag time between updates. We’ve cut those months off of that timeline. Still, a couple of months is a long time if a driver, taxi driver, or delivery driver needs an exact and efficient route. And we took that as a chance to get our hands on as much relevant information as fast as we could so that we could get to the bottom of things.