Uber Company believes that its unmanned taxis will change the way of moving millions of people. But robomobili are still very far from driving on the roads.



I stand next to a large warehouse in Pittsburgh, on the site, which stretches along the Allegheny River, where there used to be many factories and plants, and now there are shops and restaurants. I am waiting for the arrival of a new technological revolution. I check the phone, look up, and see that it has already arrived. This is a white Ford Fusion with a roof laden with futuristic-looking sensors. Two sit in front - one checks with the computer, the other - behind the wheel. But the machine controls everything. I sit down to them, press a button on the touch screen, and lean back on the seat while the Uber robot is rolling me.



Having left on the road towards the suburbs, the car clearly remains in its row, deftly slipping between the approaching car and the trucks parked on the side of the road. I was already in robo-mobility, but it is still amazing to watch from the back seat, how the steering wheel and pedals move independently, reacting to traffic events.

')

So far, most automatic cars have been tested on highways somewhere in California, Nevada and Texas. In Pittsburgh, the same winding roads, a huge number of bridges, difficult intersections, and is full of snow, rain and sleet. As one Uber director said, if a romo mobile can drive in Pittsburgh, it will be able to drive anywhere. And as if to test his theory, when we turn onto a busy street near the market, two pedestrians jump out onto the road in front of us. The car smoothly stops ahead of time, waits, and then continues to move.







On the screen in front of the rear seat a strange world is visible through the eyes of a computer - the surrounding is depicted in bright colors and sharp edges. The picture is obtained thanks to an amazing set of tools placed on the machine. As many as seven lasers, including a large rotating lidar on the roof; 20 cameras; high-precision GPS; several ultrasonic sensors. On the screen in the car, the road looks blue, buildings and other cars are red, yellow and green, and pedestrians are highlighted with something like a lasso. The screen also shows how the car brakes and taxis, and there is a button to stop. Since it's 2016, the screen even has the ability to make selfies. Shortly after the trip, a cycled GIF comes to my mail, where you can see the world through the eyes of the car and my smiling face in the corner. People on the sidewalks wave to us when we stand at the traffic lights, and the comrade driving the pickup behind us shows our thumbs.



My trip is part of the most sophisticated robots test today. Uber has already given the go-ahead for selected customers to order automatic taxis and ride them around Pittsburgh. The company, which has already turned over the entire taxi industry, allowing you to call a car with the help of a smartphone application, already in a few years is going to make the most of its fleet of cars automatic. It is a bold bet that technology is already ready to change the way people travel for millions. But in a sense, this bet is already necessary. Uber for the first half of the year has already lost $ 1.27 billion, mainly due to payments to drivers. Robomobils are “a great opportunity for Uber,” says David Kate, an assistant professor at MIT who studies innovations in the auto industry, “but there is a danger that someone will outrun them.”







Most automakers, especially Tesla Motors, Audi, Mercedes-Benz, Volvo and General Motors, as well as Google and, allegedly, Apple, are already testing robomobili. Tesla already drive independently in different situations (although the company warns that the system is better to use only on the highway, while keeping hands on the steering wheel). But despite serious rivals, Uber has the best opportunities for the rapid commercialization of this technology. Unlike Ford and GM, it can limit trips to those routes, which will initially be within the power of a computer. And unlike Google and Apple, it already has a large network of taxis, which can be gradually transferred to automatics.



The advantages of the technology are obvious to the management of the company. Without drivers to share with, Uber can make a profit. Robotics can become so cheap and simple that people no longer need to buy their cars. As a result, automatic cars can change the whole essence of transport. Uber is already experimenting with food delivery, and recently she bought Otto, a startup developing automatic systems for trucks. Robots and robotic vans can deliver food from warehouses to homes and offices quickly and efficiently. Shortly before my trip, Andrew Lewandowski, head of the robility division at Uber, a veteran of the robility management program at Google, and one of the co-founders of Otto, said: "I believe that this is the most important thing that computers can do in the next 10 years ".



Uber is developing fast. The company created an advanced technology center, where it develops mobiles, in February 2015. She hired researchers from the Robotics Department of Carnegie Mellon University (CMU), located nearby. With their help, Uber developed robotaksi in just over a year - approximately the time it takes for an ordinary automaker to develop a new design for an entertainment console.



But is it developing too fast? Is technology ready?

Predecessor robots

The rest of the time I spent in Pittsburgh, I drove a Uber taxi driven by people. The contrast was felt. I wanted to visit the State Center for Robotics (NREC) - part of the Institute of Robotics, working within the framework of Carnegie Melon University. This is one of the research groups of robotics, located at the forefront of this area. I called a taxi, and Brian took the call, arriving at the dead Hyundai Sonata. Brian says he saw several Uber automatic cars, but noticed that riding them is not as good as with him. Immediately after that, he turned the wrong way and got lost. True, in the stream, he drives as well as a robomobil. And when his navigator led us to a bridge that was closed for repairs, he simply asked the workers where the detour was, and took another route. He also offered me a friendly suggestion that I should not overpay for the hook and offer me a beer as an apology for the inconvenience. In such cases, it becomes clear that automatic taxis will give you a completely different feeling. Less wrong turns and annoying drivers - yes, but no one will help you drag the suitcase into the trunk and not return the forgotten iPhone.







I refused the beer, said goodbye to Brian and arrived at the NREC warehouse, 20 minutes late. This building is filled with amazing prototype robots. If you look closely, you can see the ancestors of modern robomobiles. Immediately after entering the Terregator stands, a six-wheeled robot the size of a refrigerator, with a ring of sensors at the top. In 1984, he was among the first robots capable of moving independently outside the laboratory, and rode around the campus of the university at a speed of several kilometers per hour. It was followed by a converted NavLab van, one of the first cars fully computer controlled. In front of the entrance is a converted Chevy Tahoe, full of computers and decorated with a device surprisingly reminiscent of the early version of sensors used on Uber cars. In 2007, this robot named Boss won a prize in a competition sponsored by DARPA. It was a significant moment for robotic, proving that they can move on a regular road among the cars. Just a few years later, Google began testing robomobils on the roads.



This trinity demonstrates how progress has progressed slowly until recently. Iron and software developed, but the system hardly understood the world from the point of view of the driver, in all its complexity and strangeness. At NREC, I met with William “Whittaker Red” [William “Red” Whittaker], a university professor who led the development of Terregator, the first version of NavLab and Boss. Whitaker says that the emergence of a new Uber service does not mean that the technology has been worked out to the ideal. “The task, of course, is not solved,” he says. “The problems of various extreme situations have not been resolved.”



And there are plenty of extreme situations. Sensors may go blind or perform poorly in bad weather, bright light or in the presence of obstacles. Software and hardware can fail. Extreme situations include handling unknown in advance events. You can not program the machine for any imaginable case, and at some point you just have to believe that she will be able to cope with any situation, using the available "intelligence". But it is hard to believe in it, especially when the slightest mistake, for example, mixing up a paper bag on the road with a stone, can lead to dangerous situations.



Recently, progress is really accelerating. Breakthroughs in the field of computer vision and machine learning give more robotic mobile video processing capabilities. If you feed a sufficient number of examples to such a system, it will learn not only to see the obstacle, but also to recognize in it a pedestrian, a cyclist or a goose strayed from the pack.



But extreme situations have not been canceled. The director of NREC is Herman Herman, a robotics who grew up in Indonesia, studied at CMU and developed automatic machines for protection, mining and agriculture. He believes that robomobils will appear, but he has practical notes on the Uber plan. “If your web browser or computer crashes and freezes, it’s annoying, but not a big problem,” he says. - And if you take the six-lane highway, in the middle of which a rover mobile is driving, suddenly decided to turn left - you can imagine what will happen. Enough one wrong command steering. "







Another problem is scaling technology. A few robo mobiles on the road are easy. But what about tens and hundreds? Laser scanners, according to Herman, can interfere with each other, and if all these cars were connected to the cloud, it would require a huge bandwidth. Such a banality, like dirt on the sensor, is already a problem. “And the most serious thing - and the importance of this field of research for us is growing - how to test and test the autonomous system for security,” says Herman.

Learning to drive

For a look from the inside, I headed across the city to talk with people who were directly developing robomobilis. I met with Raj Rajkumar [Raj Rajkumar] at the Faculty of Robotics at CMU, the head of a GM-sponsored laboratory. Rajkumar looks like a man of the old school, in the rapidly changing world of robo-mobile studies, which is often dominated by representatives of Silicon Valley. He greets me in the office, dressed in a gray suit, and leads to the underground garage, in which he works on a prototype Cadillac. In the car a lot of sensors, similar to those in Uber. But they are all small and hidden, so the car looks ordinary. Rajkumar is proud of the progress in the practical use of robotic vehicles, but warns that Uber may be too optimistic in his hopes. “It will take a long time until the driver can be completely excluded,” he says. “I think the expectations should be kept.”



In addition to the reliability of the car software, Rajkumar is worried about the possibility of hacking the car. “In Nice, a terrorist on a truck crushed hundreds of people. Imagine if there was no driver in the car, ”he says. Uber is said to be serious about this issue. They recently hired two well-known auto computer security experts. Rajkumar also warns that a fundamental breakthrough should happen in order to more intelligently interpret information about the world around computers. “People understand the situation,” he says. - We are cognitive and reasonable. We understand, we draw conclusions and react. Automatic machines are simply programmed to do this and that according to certain schemes. ”



In other words, the colorful picture I saw in the back seat of Uber represents a simplified and inhuman way of understanding the world. It shows where objects are located, sometimes to the nearest centimeter, but does not understand what these objects are and what they can do. And this is more important than it seems. A typical example is when one sees a toy on the road, a person realizes that a child may appear not far away. “The problem is that Uber makes the most money in cities and suburbs,” Rajkumar says. “It is there that unforeseen situations happen more often.”





The silver button starts the automatic mode and the big red button stops the car.



And all the problems that may occur in the experiments of Uber, can affect the entire industry. The first resonant crash with Tesla, colliding with a truck, going on autopilot, has already raised worries about security issues. Hasty introduction of any technology, even aimed at increasing road safety, can have the opposite effect. “Although Uber did a great job and advertises it as a breakthrough, it is still very far away,” says KIT from MIT. - New technologies are based on positive feedback, creating acceptance of their users, but the opposite may happen. If terrible accidents are associated with this technology, and regulators will bend, this will obviously diminish enthusiasm. ”



I personally felt the limitations of this technology halfway through the Uber car, shortly after I was invited to the driver's seat. I pressed the button to activate the automatic driving system, and they told me that it could be turned off at any time simply by turning the steering wheel, pushing the pedal or pressing the big red button. The car was going fine, as before, but I noticed how an engineer who was sitting next to me became nervous. And when we were standing on the bridge in a traffic jam, and cars were driving in the oncoming lane, our car began to slowly turn the steering wheel to the left and protrude into the oncoming lane. “Grab the steering wheel!” Shouted the engineer.



Maybe it was a software bug, or the auto sensors got confused because of the large open spaces on either side of the bridge. In any case, I quickly complied.