Mapbox moves toward maps for machines

Mapbox has signaled the next step in its strategy, with a series of announcements at its developer conference, Mapbox Locate, held in San Francisco on 30-31 May. Highlighting its 1.1 million-strong developer base and the fact that 350 million users touch its maps every day, Mapbox’s CEO Eric Gunderson announced at the conference the need for maps for machines, to enable autonomous driving, drones and other use cases.

At the heart of its announcement is the new Vision SDK, aiming to bridge the phone, the camera and the car to give developers greater control over the driving experience and the ability to insert heads-up displays directly into their apps. A neural net will run directly on the device, for real-time environment segmentation and identifying discrete features like pedestrians. The SDK also supports AR-powered navigation.

Mapbox is partnering with Microsoft to use its Azure IoT Edge Runtime, as a way to reduce latency. Another partner for the Vision SDK will be ARM, which will help provide low-power optimization to ensure the SDK can run on a wide array of machines and devices.

It also announced Intel’s Mobileye as a customer for its HD Vector Tiles maps. Effectively, these maps will give Mobileye’s autonomous vehicles a high-definition view of 200 meters ahead, with data for localization and redundant perception. Traditional latitude, longitude and altitude coordinates will be represented as sequence instructions, saving space and allowing for larger area coverage. The first European auto OEM partner to use the HD Vector Tile maps is expected to start shipping in 2019.

Autonomous driving was heavily represented at the show, with a panel discussion on the topic on the first day and a DIY autonomous car competition held on the second day. The panel, which featured speakers from Nvidia, Nauto and Starsky, mainly focused on the question of safety, how to define it and what level of safety OEMs should be aiming for. While autonomous cars are held to be safer per mile than human drivers, there’s still room for the auto industry to improve its response when accidents do happen, and the agreement generally was that the aviation industry does this well, thanks to data sharing and the practice of picking apart each incident carefully to determine what went wrong.

On a lower scale, there were also discussions from Aira, a startup that uses augmented reality headsets to help blind and vision-impaired people navigate the world with help from a human agent, and from a number of National Geographic Explorers fellows, who have used maps and location to illuminate subjects like supply chains for aquariums and undersea exploration.