A few months ago I conducted an interview with, perhaps, one of the entrepreneurs, and he is Spanish, who has the most knowledge in the world of autonomous driving. For most of those who read these lines, the name of Andrés Torrubia, surely, is not very familiar, although for the scientific circle his last name does sound, and with force. It is also known for being the terror of Chinese entrepreneurs. In his own territory, in his own language, which he acknowledges that he does not understand, and with its rules, he was able to win one of the most prestigious artificial intelligence (AI) and autonomous driving competitions organized by the giant Alibaba. And in 2017, in Silicon Valley, he was also able to rub shoulders with other Chinese, Russian, American entrepreneurs and more than 2,000 participants to finish fourth in another competition in which he had to solve a problem of perception of the environment for autonomous cars.
The funniest thing, and he tells it himself, is that in the competition that he won his nom de guerre in China in the same era ‘Sanchinarro’. A colleague and friend lives in that neighborhood of Madrid, and Andrés Torrubia could not think of anything else than to put such a name on his badge or accreditation. There, when he was already in full swing, he learned that in China Sanchinarro means “The wise sacred stone.” And good luck to you.
However, he does not consider himself a sage, quite the opposite. After passing through CERN, the European Organization for Nuclear Research, he decided to enter the field of artificial intelligence and autonomous driving, a field that is, as he himself says, in an incipient phase that will provide opportunities for everyone, and That even Spain could take advantage, for example, to remain one of the leaders in the tourism sector.
“There are experts”, says Andrés Torrubia, “who say that 80% of the problem of autonomous driving is what is called AI applied to driving. This is nothing more than making the computer in the car capable of perceiving the environment in numerical terms to be able to decide what to do. If you compare it to a video game, when you play Grand Theft Auto on the PS4 the cars in the game itself always win and never crash. One of the reasons why they are not wrong is that in the video game it is not necessary to process the data to have a numerical representation of what is around. In a real autonomous car there is the environment and you need a sensor to measure it, be it a camera, a LIDAR, a RADAR, an ultrasound, whatever, and take that data in real time and process it so that they tell us if there is a car one meter, for example. That is what is called perception, and until now it was the main obstacle ”.
80% perception
Sebastian Thurn, who is a Stanford researcher and founder of Google X, the forerunner of Google’s autonomous car, and then Udacity, says that autonomous driving is 80% perception and 20% the rest. As of today, we are still solving the first 80% of the perception. When we see that there are accidents, Tesla, Uber … they are motivated by incorrect perception.
All the projects underway today in autonomous driving use sensors. For an economic reason, Tesla uses several cameras and a RADAR, which is cheaper than LIDAR, and with these sensors its vision of the environment is made. Uber, like Waymo and others, use LIDAR, which is a RADAR sensor but with a laser that gives a three-dimensional representation of the environment. Why don’t we see LIDAR in Tesla cars? Torrubia wonders. And he answers: “because it is very expensive.” Waymo cars have a LIDAR, and if this car were commercial the price would range between $ 80,000 and $ 100,000 just the LIDAR, more than the car. But no sensor gives the information intelligently and tells you that what is in front of it is a car that occupies a certain space and volume. Once you have the information from the sensor, you have to introduce it into a system, today deep neural networks are used, which process the relevant information that has reached it, if the road is in one place and the other car is in another place.
Autonomous driving faults
“And when there are errors in perception, it is either because the sensor is not perfect or because the information is confusing for the system to analyze it. This is what happened with the Uber accident. I give you another example. Tesla cars work with cameras. Imagine that you are going with a Tesla and in front of it there is a bus that has an advertisement painted on the back of a car that is going in the opposite direction to yours. If that car, the Tesla, is going with only one camera, you may think there is a car coming from the opposite direction. And it would slow down, with what would be a tremendous failure of perception. It is a case that could perfectly happen. LIDAR is clearly a great help because it is not dependent on light and can be combined with cameras. Humans fundamentally drive with our eyes. Putting many cameras in a car to mimic the stereo vision of the eyes is Tesla’s gamble. I personally believe that it will be possible to drive at the perception level with cameras, but the market is reached earlier with LIDAR, although it is much more expensive ”, says the Spanish entrepreneur.
When will all the autonomous cars be? And how will the coexistence between autonomous cars, of any level, and with a driver? Torrubia believes that the best solution would be for all cars to be autonomous. Everything would be infinitely simpler. If they were all autonomous, they could talk to each other always functioning as a network. In China they are beginning to talk to make roads only for autonomous cars, in which there would not even be a need for signs. And the cars could go much faster. When will we see coexistence? “I am of the opinion that when the answer to a question of time is beyond five years it is that we have no idea. Yes, I can say that we will not see a 100% autonomous car in five years or by chance. My son is 14 years old, and I’m sure he will get his driver’s license, unfortunately. My grandson? Probably not”.
