The work of the future? The cartographer (for self-driving cars)
According to many experts, the introduction of self-driving cars is only five years away. By 2022, many markets will be ready for the sale of what are now prototypes (even if at an advanced testing stage). While in 10-15 years self-driving cars will be a common form of transport for millions of people. To achieve this objective the technological giants are working together with various car manufacturers. Some, such as Tesla, combine the two dimensions in a single entity with their own software and hardware Elon Musk in the same vehicle. But to go where?
The question arises naturally if you consider that it was in fact a Tesla model X vehicle that crashed last 23 March against a barrier of an American freeway in the vicinity of Mountain View, California, while the self-driving system was active. The impact caused the death of the driver and rekindled the debate on the subject of safety. And yet the stretch of road where the accident happened had already been covered more than 85 thousand times, with a total of about 200 journeys per day, by owners of Tesla vehicles (which use a machine learning system spread throughout the fleet) and the route would not have concealed any dangers for the software controlling the steering. Something certainly went wrong in spite of a system of ultrasound sensors, TV cameras and radar. And the question arises naturally: it wouldn’t be that we need better maps? Those for example that show where a white line between lanes becomes a cement wall?
Questions that put an ancient science in the spotlight: cartography. Adapted obviously to the needs of the 21st century. From the time of the Pythagoreans who conjectured, on the basis of empirical observations, that the Earth was round and then the maps that outlined the coast of the New World discovered by Christopher Columbus and more recently the details provided by Google Maps, the discipline has evolved considerably and is now at the centre of a new golden age. This is what was stated in a report of the investment bank Goldman Sachs which is why the market of maps produced for use by self-driving cars will reach 24.5 billion dollars by 2050. But from drones for delivery to real time advertising, the applications and benefits of digital cartography are immense. So much so that in the USA in the last ten years the percentage of degrees in cartography has grown by 40% only partly satisfying a demand for professionals that will grow by 30% between now and 2024. And if the percentages still don’t make it clear, just think that a colossus such as Google (whose Maps service is used on average by a billion users per month) has more than one thousand employees and six thousand contractors assigned to digital maps.
The market of maps produced for use by self-driving cars will reach 24.5 billion dollars by 2050.
Cartographers today no longer use compasses, rulers and graph paper but an endless supply of data. First of all, the open source archives such as the British OpenStreetMap and OpenAddress, already widely used by various companies to create their services. Secondly, the photos that contain various details on streets, roads and routes. A similar database is what Google fed into an artificial intelligence in 2017 starting with photos taken for the StreetView service: more than 80 billion photos by which house numbers, addresses and commercial signs were identified automatically. Lastly, the Gps data which can be obtained from the mobile devices that all of us have. Information that Mapbox, a company based in San Francisco, obtains whenever a user connects to the maps produced starting with a software development kit which Mapbox itself conceived and supplied to digital cartographers. Information which, according to the company, would make it possible to redraw the map of the bay of San Francisco at least ten times a day.
In the last ten years the percentage of degrees in cartography has grown by 40%, only partly satisfying a demand for professionals that will grow by 30% between now and 2024.
But how much does it cost to produce a digital map? We don’t have a figure but if you remember that many city maps produced starting with an aerial photograph of an area that is first processed by recognition programme and then corrected manually to remedy the differences from reality due to possible obstruction such as trees and shadows, you begin to understand that the work can be long and costly – and not only in economic terms. However, something that could think about lowering costs is the new artificial intelligence thought up by MIT: RoadTracer. The starting point is the photo of a known location in the road network. Using a neural network the AI gradually reconstructs the entire map as if it were a puzzle from time to time trying to add the missing piece. In the case of New York, Road Tracer managed to map 44% of the existing road intersections. A figure which is much better than the 19% obtained with normal computerised systems. And considering that not even the experts of Google were able to map the greater part of the more than 20 million miles of roads in the world, we are only just starting to produce the best map possible.