Dmitry Polischuk, Yandex: “It is absolutely guaranteed that unmanned taxi will be cheaper than the usual one”
The head of Unmanned Cars Department at Yandex on blurry prospects of clumsy Russian laws, key “brakes” on technology testing and substitution of taxi drivers for unmanned vehicles
By the end of the year, Yandex will launch about 100 unmanned cars on public roads in Moscow and Tatarstan. The experiment has already been carried out for six months, underperforming Russian legislation and bureaucracy are the main barriers to technology development, head of Unmanned Cars Department at Yandex Dmitry Polischuk claimed. Last weekend he visited Tatarstan to test a completely autonomous car as a wedding motorcade. Realnoe Vremya’s correspondent managed to talk with him and find out why Yandex didn't rush to monetise its developments and patiently waited for Russian legislation mature and simultaneously looking at “friendlier” markets.
“There aren’t standards and norms for unmanned vehicles in general, the existing norms for usual cars are tried to be tailored for unmanned vehicles”
Dmitry, according to my subjective observations, the level of people’s understanding of what is going on with unmanned vehicles now is probably compared with the level of understanding of 5G and its real prospects in the next five years — everyone has heard about it, has some idea about it but fails to understand what’s the gist and when this all will happen. A three-year experiment on testing of unmanned cars on urban roads in Moscow and Tatarstan kicked off on 1 December 2018, at least it was announced. What interim results do you have? Is there any advance?
The decree you are talking about is applied precisely to the testing of the technology in big cities. If we have a look at the main locations we are testing the unmanned vehicles on public roads now, it is mainly Moscow, of course. The decree finally regulated tests of unmanned vehicles in a big city with autopilot on but with a driver at the wheel who is responsible for everything and controls the whole process.
This story was launched in Innopolis a bit earlier — late last summer. Even the unmanned taxi has been operating here for more than a year. And unlike many analogous stories, it operates here in a more authentic format. If we look at what other companies that deal with the same issue in the world we do with say, we can see that analogous unmanned taxi services also operate in other countries, but a person is at the wheel in the car — a safety engineer just in case. Here, in Innopolis, a person is still in the unmanned car but in the passenger seat. The environment, difficulty of tasks — this allows doing it this way precisely here. Actually, it is an absolutely unique story for Europe, it isn’t frequent around in the world, such a level of technology can be rarely found.
This story was launched in Innopolis a bit earlier — late last summer. Even the unmanned taxi has been operating here for more than a year. And unlike many analogous stories, it operates here in a more authentic format
Back to the story about legislation and this decree, which is quite interesting and complicated, the problem is the following: technologies outstrip the law. Nobody completely understands what should be done. When one starts to talk about unmanned vehicles, everyone who doesn’t deal with this issue directly finds it hard to understand how it actually works: to what one should pay attention and to what not. I will put an example. The first thing we face in the testing is a big number of problems of certification of these cars. The car must go through this procedure to appear on the road. But regulators pay huge attention to the fact that a car is a car and check the brakes, the parking brake, indicators, lights, while all this doesn’t have a lot to do with unmanned technology.
Don’t they check the “brain”, do they?
It is impossible to check it. All that can be done nowadays within current legislation is to verify cars according to known established norms. In other words, to establish that mechanically it is a car as well — one can drive it, put on the brakes, it has wheels, they turn. The existing certification of the unmanned vehicle is a transport certification, what is done when a new car enters the market to permit its sale, events close to MRO.
As there aren’t standards and norms for unmanned vehicles in general, the existing norms for usual cars are tried to be tailored for unmanned vehicles. In other words, on the one hand, during the certification, indeed, there was examination, the car is in a good technical state and it can be used, but in our opinion, all this is quite excessive and doesn’t add security but only scotches the process of delivery of unmanned cars to the road, their testing. This is what we try to do at this stage: we ask permission to test in a mode when the driver keeps control of the car, and at any moment we will say if the car can’t handle the situation on its own, the person can take control. And it is the key moment.
“By the end of the year, we hope to increase the number of unmanned vehicles on roads to 100 cars”
Isn’t it already done this way thanks to the same decree?
It’s done this way, but you don’t imagine how much time it took us and how many problems we face because people don’t understand. Why such a test matters, nothing we do in closed testing areas, simulators is enough to make sure that the technology really works. Use on public roads is the only final check digit.
Yandex unmanned cars have been driving in Moscow for quite a long time, since as early as late 2017. But they didn’t use to drive with the autopilot on but with sensors, collected data
As I understood, a month and a half ago, on 20 June 2019, Yandex began full-fledged testing of unmanned cars precisely in the biggest metropolis, Moscow, right on public roads.
No, here it is important to understand what is considered as a test. Yandex unmanned cars have been driving in Moscow for quite a long time, since as early as late 2017. But they didn’t use to drive with the autopilot on but with sensors, collected data. Now they have got a chance to test with autopilot on under human’s control.
I should note an important moment: the problem is not that it couldn’t be done but that it was a grey area. The very fact if trips with autopilot on in the car equipped with sensors and autonomous management system were permitted or not was a grey area. Now, with the appearance of the decree, this is clearly regulated: yes, there is such a format, yes, one can drive it, yes, a person must be at the wheel and he is responsible for everything that’s happening to the car. Consequently, since the decree appeared, we’ve had a chance to switch the cars to a completely autonomous mode.
According to Yandex’s latest statements, it is clear that forced certification of every sample of the unmanned car before its launch on the big road was the main obstacle. The government talked about the work on simplification of the certification. Is the situation changing? How many unmanned vehicles are tested on roads?
Yes, as I said, the problem was that we even had so many cars that were able to do it during the start of the testing, in December 2018, while we didn’t manage to get permission to almost use them with autopilot for a long time because of slow certification. Now the situation has changed: tens of cars are driving in this format in Moscow now, more than 30, and we actively keep increasing the autopilot mode. It is important that by the end of the year we hope to increase the number of unmanned vehicles on roads to 100 cars. And at the moment we are carrying out the plan.
Are the hundred unmanned cars for Moscow only or is it considered to increase the fleet in Tatarstan as well? There were talks about the expansion of the test zone, the inclusion of other regions.
Actually, it doesn’t matter. All those zones where the decree permits testing at this moment are theoretically available for us. It is Moscow and Tatarstan. Now a supplement to the decree is being prepared, it will expand the list of regions, but, as I already said, not region but the possibility of the procedure, elaboration, acceleration itself matters. If it happens like it used to be, the certification of one car will take a month, it will take 4,5 years to certify the 100 cars. Now the process has changed for the better, and it seems more realistic to launch big volumes of cars on roads.
Here it is important to understand why we need these volumes: it might seem that your car just drives, you are welcome to test it. The problem is that the volumes are needed to verify the technology. Now the testing is mainly focused on measuring the level and amount of human’s interference in the operation of autopilot, and this amount needs to be reduced to develop the technology. To have an objective indicator of quality, cars have to cover many kilometres autonomously. And we will be covering these kilometres with a small fleet. We have to have tens, hundreds, probably even thousands of cars to quickly verify that the technology will remain safe after changing algorithms when some things are added to them.
There is a traffic code, it is quite strict. And if we imagine a utopian picture where everyone follows the traffic code, it is hard to drive on roads
“Other people became the main problem when testing the unmanned cars”
Can we lift the veil a little bit? How are the cars tested? What problems do developers have today?
Other people, that’s to say, other road users — cars, pedestrians, cyclists and so on — became the main problem when testing the unmanned cars. There is a traffic code, it is quite strict. And if we imagine a utopian picture where everyone follows the traffic code, it is hard to drive on the roads. As a rule, an unmanned vehicle has difficulties because of two things. Firstly, not perfect road infrastructure — road repairs, which sometimes make the car to pull to the side of the road to go through.
The car can be taught whatever, but it takes time. We can teach it to avoid on the shoulder if there is a repair, but again, there is a human. Road signs must be installed when there is a repair, they must be installed correctly and in advance, while they are usually installed haphazardly, people do everything by breaking rules, standards and regulations a bit. Secondly, people are the human factor. To hell with road repair works, everyone violates the traffic code, the speed limit, ignore the road surface marking, change lanes wrongly, don’t indicate manoeuvres…
But the computer identifies objects, both living and non-living, it seems that it can also identify and analyse their moves…
The whole process of the unmanned car’s movement is based on four basic things. Firstly, it understands where it is. Secondly, to understand what is around it, to find people, cyclists, bikers, the marking, signs and so on. And the most interesting phase comes — the car has to predict what will happen to all this later. And here the human factor is more uncertain. If everyone strictly followed the rules, it could be quite easy to predict. But the typical situation is when someone needs to turn to the left, he is on the far right lane and starts moving across all the markings at the last moment. In other words, the unmanned vehicle has to presuppose at any moment that somebody can move as it is comfortable for him, not as it should be, it has to consider a huge number of variations.
It seems that you should have started with India or Thailand, after the Asian traffic you wouldn’t have had problems in Russia.
Yes, there can be illustrative cases when there is a lot of two-wheelers and they move completely chaotically, but the introduction of technologies starts from some simple, structured places towards more complex places, the unmanned vehicle has to master simple conditions first.
“The Yandex car is more ‘polite’ in Innopolis than the one in Moscow”
I want to go back to Innopolis. Today we’ve witnessed the use of an unmanned car as a wedding motorcade, moreover, the car moved completely autonomously without an engineer and human involvement, which is a unique event. We should note it didn’t happen on public roads but, in fact, on the internal area of the technopark. A petition surprised me — to stay at least two meters clear of the car’s trajectory so that it wouldn't react because of people and stop earlier than needed. As I understand, the cars in Innopolis and on Moscow roads are identical, how do they move on congested urban roads of the capital if they stop two meters far from any object?
Yes, from a perspective of the set of sensors and algorithms that are used, they are identical. But Innopolis has special settings, which changed a bit. Let’s say that the level of the car’s confidence changed. It is more “polite” in Innopolis than the one in Moscow. The reason is simple: here traffic is low, there are few cars and people. This is why the car drives here as a driver who has recently got a driving licence. First of all, it is done for safety. Though there are few cars here, there are quite interesting situations here, including with lawbreakers: when the bus goes along the avenue facing the oncoming traffic or moves along the ring counterclockwise. I can’t say it happens everywhere here, but it happens. A tractor going on the oncoming traffic lane would cause righteous anger in a big city, here it is alone…
Today it was important for the shooting that the car stop in the place we had agreed on, this is why the car was very polite, and this is why you had to stay clear. Because the situation when a crowd of people surrounds the car is very untypical. Of course, it can also be taught to drive in a crowd — to carefully crawl, look for loopholes and so on. The question is why deal with this if 99,9% of situations on the road are different: people walk on the pavement. Yes, they go down, sometimes cross the road in the wrong place and so on, one should be ready for it, but nobody surrounds the car. We will handle these situations sooner or later, the problem is we should do it consecutively.
Last year, Israel permitted Yandex to test unmanned cars in Tel Aviv. If I am not mistaken, it is the first country after Russia and the USA where your unmanned car appeared in the city’s streets. Where is the most development legislative base to introduce the technology of unmanned vehicles?
We have tested it in three countries at the moment. The latest legislation for unmanned cars is in the USA now. As I think because precisely this part is regulated in every state, not by federal legislation. There are states that compete with each other, which allows promoting the law faster. For instance, California made up a rule to test, a number of companies put their cars on the road, began to test weakly, then Arizona came and said: “Everything is very simple here, you do it this way, and that’s it”. Everyone rushed there immediately. It is competition for investments at the level of legislation.
Secondly, improvements in cars are regulated much simpler there. In Russia, you hammer an extra nail, you must certify it. In America, you can see pick-up trucks with horns on roads — in Russia, such cars can’t be certified at all. The next moment: plenty of companies dealing with unmanned technologies are based in the USA — there is good weather, good demand. All this creates an impulse for lawmakers.
As strange as it might sound, in Israel we were first to begin to develop this topic. In general, the culture of IT start-ups is very strong there. Many IT companies have their own technological hubs. And dealing with such an issue as unmanned vehicles, one should enter markets that have staff besides other things. When we explored where we could start the testing, what and how to test outside the domestic market, of course, we wanted to understand if it was possible to change current legislation to make it official. In fact, this process took place when we brought the first car, showed that it ran safely, how it worked and so on. Now we are also expanding the test.
Russia is falling behind, to be honest. A more favourable climate could be created to develop such technologies. This process of certification, allowance of the car improved with sensors and other things to roads still takes much time. Even such a simple thing is now tough, slow and hardly develops. What if we go further? We want to put the engineer to the passenger seat first and then take the person out of the car. And all this will have to be approved by legislation. At the moment we stumble in the simplest step… Now everything is very slow.
If everything is calculated and designed for three years, it turns out that the price of one unmanned prototype is already cheaper than the use of one taxi car with a salary of all drivers if the person is taken out from there
“Now one prototype costs about 6,5 million rubles, we began with 9,5 million”
Is it possible to forecast the date of launch of full-fledged autopilot car on roads?
It is very hard to say. If we move at the same pace we have now, we will hopelessly fall behind. We are a company that actively invests in technologies, of course, we use the markets that welcome a) the testing, b) the implementation of these technologies. What’s the sense in knocking the closed door…
What’s the price of one prototype, its testing?
To test a car now, you need to have a person in. Consequently, there are costs on the person’s salary, if you want him to drive 24/7, you have to organise a 24/7 schedule. It is when it comes to the process itself. Besides, there is car assembly. Now one prototype costs about 6,5 million rubles. And it is important to understand this number has reduced in the last two years — we began with 9,5 million, now it is 6,5. We’ve reduced it because sponsors have become cheaper, by means of volumes and production optimisation. And it will keep cheapening. By the way, it is another misperception that an unmanned car is too expensive and unreal.
Everyone reads the news about Musk who has been promising to have at least zero costs for three years already.
No, I will say a serious thing here. If you take the price of the taxi car driving for 24 hours a day… To enable it to drive for 24 hours a day, you have to add the price of the driver, even several drivers who use this car. If everything is calculated and designed for three years, it turns out that the price of one unmanned prototype is already cheaper than the use of one taxi car with a salary of all drivers if the person is taken out of there.
Have you calculated it?
How much cheaper?
It doesn’t matter because the orders are already coinciding when we are at the testing stage. And if we imagine the situation in 2-3 years considering how cheap everything is becoming… This all absolutely guarantees that unmanned taxi will be cheaper than the usual one.