How to Build a Map for a Self-Driving Car?

How can you create a map of every road in the United States that would show every stop sign, land marking and traffic light – and make it update to every road repair and accident in real time? This question is vital for car producers and IT companies that are working on self-driving cars development. One day you would be able to sit back and read a book on your way to work while the smart car would get you there safe and sound.

Some believe that highly detailed three-dimensional maps, which would understand car surroundings and pinpoint its location, are not a necessity. All the discussions are now focused on several sensors that allow a car to perceive the obstacles on the road – these include radars, cameras and laser-based radar known as lidar. But the digital map is a bigger piece of autonomous car puzzle.

Chief technology officer at Mobileye (Israeli company that works on vision systems for automobiles) claims that if we want to create an autonomous car driving everywhere, we should have virtual maps everywhere as well. However, creating such map is quite a task in itself – 4 million miles of roads in the US alone and perpetually changing road situation are the biggest challenges. Google’s former self-driving car division, which is now a separate Waymo company, has created maps for several roads around the US. Yet rounding up the entire project via Waymo would be too costly.


The solution came from the Internet and now it seems to be the best – it’s crowdsourcing. Outstandingly high percent of cars driving around the US with Mobileye – the system that detects obstacles on the road – would help to gather all pictures and locations into a giant digital map of the entire country. Starting from 2018, BMW and Volkswagen would do that for all their new cars, and soon other carmakers would do the same.

Teaching the map to adapt to the current road situation is another challenge. A tech startup called Civil Maps is working on the issue and even managed to attract some investments from Ford Motors. Raj Nair from Ford Motor has vowed to start the production of fully autonomous cars with no steering wheels and pedals by 2021. However, to make it possible, they need a high-definition map with all the fixed objects marked on it. It is presumed that all the data maps would be assembled into a Road Book that would be sold to autonomous car owners.

Despite all the confidence, the most advanced car radars and cameras cannot help a car navigate in the hectic and chaotic world safely enough for passengers to relax and pay no attention to the road. While all those cameras and lidars help cars to see, the digital map is a “foresight” system for it. Accessing the detailed map, a car would know there is a stop sign or an overpass ahead even if its view is blocked by pedestrians or anything else.

One tragic yet illustrative accident made evident the need for a virtual map to complement the radar and the camera. In June 2016, a man from Ohio died in a car crash when he was riding in his 2015 Tesla Model S on Autopilot mode. The car was traveling at 74 mph and Tesla sensors failed to distinguish a white tractor-trailer crossing the highway against a bright sky, probably mistaken the vehicle for an overpass.

If this Tesla had a digital map to help the sensors, it would have known there is no overpass at this point of the road and would detect an obstacle that blocked the way. The following investigation of the accident and the thorough Tesla inspection did not find any defects in its systems. However, the company was suggested to update the Autopilot system and be more careful with advertising.

Tesla did follow the advice. They’ve added some primitive digital mapping. Now Tesla cars that have the new software can detect road signs, bridges and other object and transmit this information to the database, so all the Tesla cars could make more safe decisions on the road.