Researchers from the Massachusetts Institute of Technology (MIT) have developed an automated method that uses artificial intelligence (AI) to create road maps that are 45% more accurate than those made using existing approaches.

The algorithm, named RoadTracer, uses data from aerial images to train a neural network, which examines the area surrounding a known road to determine which point is most likely to be the next part on the road.

The research team, from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), say that not only is RoadTracer more accurate, it is also more cost-effective than current approaches. Having accurate and gap-free maps is particularly important for self-driving cars to operate effectively.

MIT professor Mohammad Alizadeh, a co-author on a paper about the system, said he believes RoadTracer could help map areas that are currently difficult to chart.
“RoadTracer is well-suited to map areas of the world where maps are frequently out of date, which includes both places with lower population and areas where there’s frequent construction,” he said.

“For example, existing maps for remote areas like rural Thailand are missing many roads. RoadTracer could help make them more accurate.”

Current mapping approaches train neural networks to identify individual pixels on aerial images and identify them as either ‘road’ or ‘not road’. Because aerial images can often be ambiguous and incomplete, such as shadows from trees or buildings obscuring the beginning and end of the road, the automated system can produce errors. This requires a post-processing step to fill in the gaps, which can also make incorrect assumptions, such as connecting two road segments because they are next to each other.

RoadTracer creates maps step-by-step. It starts at a known location on the road and uses a neural network to examine the surrounding area to determine which point is most likely to be the next part of the road. It then adds that point and repeats the process to gradually trace out the road one step at a time.

“Rather than making thousands of different decisions at once about whether various pixels represent parts of a road, RoadTracer focuses on the simpler problem of figuring out which direction to follow when starting from a particular spot that we know is a road,” said co-author Fayven Bastani.

“This is in many ways actually a lot closer to how we as humans construct mental models of the world around us.”

The team trained RoadTracer on aerial images of 25 cities across six countries in North America and Europe, and then evaluated its mapping abilities on 15 other cities. They found that it had an error rate 45% lower than traditional methods.

While it is more efficient, the team doesn’t believe RoadTracer would completely remove the need for human involvement.

“That said, what’s clear is that with a system like ours you could dramatically decrease the amount of tedious work that humans would have to do,” said Alizadeh.

Instead, RoadTracer’s incremental approach could make it much easier to correct errors, with human supervisors able to correct them and re-run the algorithm from where they left off.

The team is conducting separate research into algorithms that can create maps from GPS data to provide information about roads with overpasses and underpasses, with plans to eventually merge both approaches into a single system.

The paper, which will be presented in June at the Conference on Computer Vision and Pattern Recognition in Salt Lake City, Utah, is a collaboration between MIT CSAIL and the Qatar Computing Research Institute.