In order to be fully safe and autonomous, self-driving vehicles will need to monitor roads not only with cameras and sensors to decode the surrounding environment, but also with high precision maps that will allow them to travel at any time of the day and under any weather condition.

 

Map Making

The art of mapmaking or cartography has roots deep in the history of mankind: from the cave paintings in Lascaux to the ancient maps of Babylon, Greece, Rome and Asian civilisations. Maps accompanied the age of discovery and allowed us to take to the roads in mass from the 1950s onwards. The maps of the future, however, will shift from two-dimensional drawings to three-dimensional data models that will only be read and understood by artificial intelligence systems.

A completely autonomous vehicle, one that can drive anywhere, at any time, under any weather conditions, will need to do more than just trust its sensors and cameras. It will also have to depend on centimetre-precise information coming from 3D maps that are being developed as you read this and that will be many magnitudes more precise than any currently available GPS navigation map.

 

High Definition Maps

Work on these high definition maps, which is a relatively new endeavour, requires fleets of vehicles equipped with state-of-the art sensors, cameras that approach 100-megapixel definition, LIDAR scanners and GPS Inertial Measurement Units that drive around creating a 3D scan of roads and their surroundings. The accuracy of these measurements is unlike anything you may have seen on the Internet to date. It will be a true digital recreation of the world or, at least, of its road networks.

Once a functional 3D map model is created the idea is to update these maps by crowdsourcing data from the vehicles that will employ them. However, in order not to overburden the vehicle artificial intelligence systems and consume too much bandwidth, these systems will only have to transmit information on changes in the road network, road conditions, closed roads, etc. collected by the vehicles’ sensors, radars and cameras.

 

Big Data and New Standards

The effort to produce and update these new maps, which will require new standards to allow their use by all types and classes of vehicles, is pushing the development of machine learning and neural networks to fully exploit the vast masses of big data that will be constantly monitored, analysed and shared by self-driving vehicle AI platforms.

This new class of maps is expected to be available in 3-5 years, in time to be used with the first true self-driving vehicles to hit our roads. And they will probably spell the end of those hard-to-fold maps we learned to read as children.

 

For further information: