MIT Artificial Intelligence 'RoadTracer' Can Map Out Roads More Efficiently Than Humans

Tuesday, 17 April 2018 - 7:23PM
Technology
Artificial Intelligence
Tuesday, 17 April 2018 - 7:23PM
MIT Artificial Intelligence 'RoadTracer' Can Map Out Roads More Efficiently Than Humans
< >
MIT CSAIL
Traditional paper maps may still exist, but in the age of smartphones, they've become something of a novelty while more people rely on apps like Google Maps to get where they're going. 

But as advanced as these apps are, they're still predominantly mapped out by humans who have to manually trace out roads, a monotonous job that leaves several chunks of roads throughout the world uncharted in the digital world. The only thing that could more accurately chart out the world's roads would, of course, be an artificial intelligence

Enter a group of researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), who've developed an AI program called RoadTracer that does just this. They've just published a new paper containing their work, and claim that RoadTracer can create road maps with 45 percent more accuracy than other methods that companies behind big map apps typically use. See it in action below:



RoadTracer creates its maps with nothing more than overhead visual photos of the region, and while some similar neural networks have attacked the problem by labeling pixels as either "road" or "not road," CSAIL sees RoadTracer as being more nuanced.

Instead, RoadTracer will start at a point it knows for sure is a road (so it needs a little information going into a new region), and then uses a neural network to analyze the surrounding area in the image and determine the mostly likely spot for a road to continue in. And it does this in a step-by-step process until a full road is completed, which would be tedious for humans, but an AI doesn't get tired.

It was tested on aerial photos of 25 cities, and ended up performing fairly well after its resulting maps were matched with existing ones. While the video above shows RoadTracer creating maps of New York and other major cities, the AI might best be used in more rural areas with out-of-date info or no digital info whatsoever.

Mohammad Alizadeh, an MIT professor and co-author on the paper, said the following in an CSAIL press release:

Opening quote
"RoadTracer is well-suited to map areas of the world where maps are frequently out of date, which includes both places with lower population and areas where there's frequent construction. For example, existing maps for remote areas like rural Thailand are missing many roads. RoadTracer could help make them more accurate."
Closing quote


If RoadTracer ends up being adopted by big digital map-makers like Google or Apple, it could lead to much more accurate maps in more rural areas.

So the next time you take a wrong turn out in the middle of nowhere, you'll still have some idea what roads you're driving on

Science
Science News
Technology
Artificial Intelligence
No