Mapbox acquires neural network startup Mapdata to help it expand into AR maps

Two weeks after announcing $164 million in funding, mapping data startup Mapbox has made an acquisition to help build out its next steps into autonomous driving and other applications that it believes represent the future of navigation. The company has acquired MapData, a artificial intelligence mapping startup based out of Minsk in Belarus that focuses on building neural networks to improve computer vision for applications like augmented reality.

It will be using the deal spearhead the opening of a new office in the city and the hiring of more engineers; and to help build its next big product, an SDK that will let developers build augmented reality-based maps into their apps that will work by way of the front-facing cameras on people’s devices. This SDK should now be coming in Q1 of next year, Mapbox CEO and founder Eric Gundersen told me in an interview.

Before the acquisition, MapData had been a partner of Mapbox’s providing engineering support on open source projects for the past 18 months, but — ironically for a mapping startup — it was largely off the grid as far as startups are concerned. (Just try finding a website for it, or much other proof that it exists.) I guess that if you’re good at helping people figure out where they are, you are probably also very good at making people invisible.

Terms of the deal are not being disclosed. Yuri Melnicheck, a Belarussian entrepreneur and investor (who was also behind the sale of AIMatter to Google this summer), is an advisor to Mapbox. Melnicheck had made the introduction between the two companies.

The CEO of MapData who will now lead Mapbox’s Belarussian office, Alexander Matveenko, had worked at another previous company of Melnicheck’s,, which had been acquired by in Russia. was a Waze-style platform that was built in part through crowdsourced data from its users. That is notable, because it’s a paradigm that Mapbox has also used in building its platform up to now, and will continue to use in the future as it ramps up its own vision for how to create and deliver navigational guidance to the world.

Today, Mapbox is already collecting anonymized (and ONLY anonymized, says Gundersen) aggregated telemetry data from 200 million users, by way of apps that are using the Mapbox SDK, which include apps from Airbnb, Instacart, Snap, and MasterCard. “When people touch our maps, we get latitude, longitude, topology and more,” Gundersen said.

Acquiring a neural network and AI specialist to help build Mapbox’s AR-based SDK will serve two purposes: it will (Mapbox hopes) be creating a service that people will use, and it will be creating an interface that will start to be able to collect even more data, specifically in the area of street views, which is essential for building out Mapbox’s dataset in its bid to do more in the area of autonomous driving systems.

“The AR SDK is focused on AI and presentation, yes, but my interest ultimately is, the more our SDK is used, the more data we get back,” Gundersen explained. “If we do this right, everyone will benefit from that network effect.”

There are still a lot of questions about how this will work, and what applications might come out of AR maps longer term.

Many in the mobile world have talked about how the phone is “disappearing”, with bezels getting thinner and more functions requiring less of your active input to work, but there is still a hurdle when it comes to AR applications on these devices since they, by definition, require active participation and physically holding up your device as you walk.

“I think it will be interesting to see how socially acceptable this is,” said Gundersen. “For walking, I think AR and holding up your phone feels awkward and I don’t know if that will ever be publicly acceptable.” 

For now, the more obvious deployments are in areas like gaming and in-car navigation — two use cases that Mapbox is doubling down on with this deal.