At the beginning of July, a video with a driverless Yandex.Rover in the streets of Moscow appeared online. The technology is supposed to become a useful tool in e-commerce - it is evident from the fact that Amazon corporation launched similar rovers in the summer of 2019.
Source: RIA Novosti
ICT.Moscow decided to learn about Yandex’s plans on implementing the technology in Russian e-commerce and asked about the peculiarities of working on driverless transport. The representative of the company revealed where the rovers have already been implemented, spoke about their readiness for real operation and explained the details of the development of a universal technology for self-driving vehicles.
Read the interview of ICT.Moscow to learn where the delivery robots will appear first, when the driverless transport technology will be ready for operation in the most difficult road conditions and where and why Yandex tests its self-driving cars.
— You started to use Yandex.Rover in the streets of Moscow. Does it mean that the technology will soon be ready for operation in real-life conditions?
— We have been using our rovers in the streets of Moscow for quite a long time for collecting data, testing and training algorithms. This time a journalist was nearby, and he shot the standard working process.
In terms of operation, we have already conducted several pilot projects and the first launch of the robot outside the Yandex ecosystem. We started to develop Rover in the summer of 2019, and already in November we conducted the first trial - the robot was delivering documents and small parcels within our headquarters. November was a good time to launch Rover because of the weather conditions: the robots had to deal with rain, snow and ice. The robot leapfrogged favorable weather conditions and faced harsh Russian realities straight away.
In February this year we carried out a test integration with “Beru” service. The piloting was linked to February 14th, when there was a high demand for delivering a large number of parcels. However, only Yandex employees could use the delivery - Rover picked up parcels at a single center and delivered them to buildings.
In April, the robot got its first serious job - it started to operate at Skolkovo. The territory of the center is arranged in the way that the buildings of different offices are situated at a certain distance from each other. Previously, the employees had to use a car to get to several places to exchange documents and correspondence. With Rover it is no longer needed - the robot took over this function. Its route there is about 2-3 km.
In fact, Skolkovo buys robot delivery service from us, so, technically, it is the first commercial integration of the technology.
Less than a year has passed since the emergence of “let’s create a robot” idea till the first commercial application.
— In other words, is the technology ready for real-life application? What are the plans for further implementation of Rover in real commercial and business processes?
— We are currently negotiating with several other companies interested in small cargo delivery using robots. We expect that Rover will fit into Yandex ecosystem, including FoodTech sector, which has become more in-demand than before due to the pandemic.
Such delivery robots allow to successfully develop e-commerce in general. This sector is rapidly growing and requires more and more people at different stages. In the USA there is a shortage of staff in the cargo transportation, because young people are reluctant to enter this profession. Such rovers are able to meet the increasing demand for water, food and goods delivery.
If we talk about the possible range of application of such rovers, there are many tasks in logistics that can be automated using these robots. They not only can drive through the streets and deliver documents, parcels and food, but also can be used within warehouses. When we announced the introduction of the robot within the company, our colleagues from data centers, which occupy large areas, showed a great interest in it. In case there is a need to deliver two ram cards from one point of the data center to another, a rover can help.
The main range of self-driving couriers use includes the services where last mile delivery is needed, primarily e-commerce and warehouse logistics.
— The operation process within data centers and warehouses in quite clear, but Rover will also be operating in the streets, with a large number of road users and unpredictable situations. To what extent is the technology ready for urban environment?
— The technologies that were perfected and tested in driverless vehicles helped us a lot. We had to rebuild some algorithms, but the basic principles are the same. It was one of the reasons why we managed to carry out Rover piloting so soon: it comprises many skills that have already been developed. Moreover, the scenarios of use are more simple, even in terms of safety: a car drives at about 60km/h, so the safety requirements are much higher.
Rovers drive on sidewalks at an average speed of 5 km/h, in bike lanes the speed can be higher (this scenario is being tested in Skolkovo). They have already learnt to accurately detect and recognize the signals of pedestrian traffic lights (in the same way as a self-driving car detects turning arrows). But currently there is a difficulty with uncontrolled pedestrian crossings, as it is not clear how Rover should behave at them - it is neither a car, nor a pedestrian. So far, the solution is as follows: the robot will let a car pass, but if it understands the driver sees it and lets the robot pass, Rover will drive across the crossing.
One of the main tasks of machine learning in self-driving transport is to learn to predict the behavior of other road users.
It is not enough to determine where they are driving or going and forecast their trajectory - intentions are more important. Especially in case with self-driving vehicles: will the pedestrian. standing at the median strip, cross the road or not?
To determine this, neural networks use all available data on the object and the environment. For example, in case with a car it will be the position in the lane, steering angle, previous acceleration, road environment, etc. Moreover, such scenarios have been tested many times in real road conditions. If in the previous 10 thousand similar episodes a person (a driver or a pedestrian) behaved in a certain way, in the new episode we can predict future behavior.
— And what about security? Rovers will have to deliver value parcels as well. Is it safe to leave the unattended?
— Firstly, they are not left unattended. There are GPS tracker and various sensors and cameras in each Rover. If the robot considers something went wrong, for example, if it deviates from the set course, it will send a signal to the operator, who will be able to access remotely and see what is going on.
Secondly, a Rover is quite heavy, so it will not be easy to carry it away. Even if someone manages to do it - what for? You do not know what is inside, and it will not be easy to pick the lock. If someone succeeds, there can be an inexpensive product from an online shop. It will also be difficult to resell the parts of Rover: it will be clear where the part came from. And again, GPS trackers are installed in Rovers, and they allow to track its movement, just like in car sharing.
There are not so many scenarios in which something threatens a rover.
— You said that the developments for self-driving car were used in Rovers. So, it is a universal technology for unmanned transport that requires a minor adaptation, right?
— Absolutely correct. Once again, Rovers drive in more simple conditions than self-driving cars, with more factors to consider during operation. We will have to pilot self-driving vehicles in different countries to create a technology that can be implemented worldwide - only Russia is not enough for this task.
There are unique things in the USA, which cannot be seen anywhere in the world. For example, a four-way stop, where the rule about the right of way, according to which the right-of-way should be yielded to the vehicle on the right, does not apply. If you arrive at the intersection first, you can continue driving. In Tel Aviv, where we also test unmanned vehicles, there are things that are not so common in Russia, but can be seen in other countries. For example, roundabouts are not very popular in Russia, but in Europe they are practically the main element of road infrastructure. In Tel Aviv there are a lot of two-wheeled vehicles: mopeds, self-balancing unicycles, scooters, bicycles, electric scooters, etc.
In Moscow and in Russia in general we also do have these things, but there are many more of them in Tel Aviv (and they are available all year round - one more advantage in for training self-driving vehicles algorithms). If we are able to drive in difficult conditions in Tel Aviv, and then will go to Italy, for instance, we will be able to drive with scooters, on roundabouts and will be prepared for other peculiarities.
It is also important to take the local driving culture into consideration. In Tel Aviv, for example, you should be very persistent to drive.
If you are polite and give way to other cars, you will not drive far.
We noticed this feature in Las Vegas. Local residents, after taking a ride in our self-driving vehicles, noted that the car changes lanes very quickly. It can be explained with the peculiarities of driving in Moscow: you have to be quick to change lanes. In the US it is easier - the people are in no hurry and will give way to you. We configured the system to drive like locals - we changed settings so the car could change lanes more smoothly. And Tel Aviv, on the contrary, teaches you to be more persistent. This experience will be useful for us in Moscow as well.
Source: Israeli TV
— V2X infrastructure can be of great help for self-driving vehicles, it allows the car to exchange information with road infrastructure objects and other vehicles. How actively is this area developing in Yandex?
— That also brings us to the issue of technology’s versatility. If we create a technology that should be safe and applicable anywhere, we should rely only on ourselves. Communications with infrastructure and other driverless vehicles is good and useful, and we are testing such scenarios. But we cannot rely on them and make them an indispensable condition.
For example, a person ordered an self-driving taxi to a location that is far away, with no infrastructure and no self-driving vehicles around. The vehicle should be able to drive safely by itself - as if there was V2X infrastructure around it.
Moreover, different countries have their infrastructure with their standards of data transmission. This problem exists, even without V2X infrastructure. For example, in the USA there are traffic lights that have eight different signals, including a yellow flashing arrow.
In general, the possibility of receiving information about the signals one kilometer before the traffic lights is a good advantage. Having this information, the vehicle can adjust the speed, plan the motion path in order to cross the intersection as efficiently as possible, with minimum fuel and time consumption. This is a nice bonus, but it is still extremely important to be able to recognize traffic lights signals using vehicle’s vision.
It is not always possible to rely on 5G in data transmission. Firstly, 5G coverage is available not everywhere. Secondly, the signal can be unstable even if the network is available. All these are additional risk factors.
Finally, imagine how much the dependence on V2X infrastructure could postpone the launch of self-driving vehicles. Then it would be necessary to re-equip all roads, which is a time-consuming and labor-intensive process and requires a lot of investment. If we want self-driving vehicles to be implemented on roads and provide services (taxi, for instance) as soon as the regulatory measures are finalized, relying on V2X is not practical.
— So, if the self-driving couriers and vehicles technology is ready for commercial use, do we have to wait only for the finalization of the Russian regulatory measures?
— In case with rovers, no additional regulation is required.
As for unmanned cars, in general, we are ready to launch the corresponding services in small cities, with not very intensive traffic. We managed to test such a service in Innopolis, where it has been operating for about two years now. The conditions similar to Innopolis are already suitable for launching such services.
As has already been mentioned, the most difficult task for self-driving vehicles is to predict intentions. The more road users there are around the vehicle - the less predictable the environment becomes.
The trajectory for the emergence of self-driving vehicles will be roughly as follows: small cities — suburbs of large cities — centers of large cities.
Rush hour in the center of Moscow is the most unpredictable time for road traffic. With the increase in the number of cars, the number of traffic rules violations also grows. The traffic in the center of major cities poses a challenge to self-driving vehicles, which needs to be solved.
Of course, now unmanned cars can drive during rush hour in Moscow safely, without creating emergency situations for other road users and themselves. But it will take them more time to drive through difficult junctions than it would take an experienced driver. In every ambiguous situation the vehicle will give way to other road users.
But when data accumulates, the situation will change. It can be compared to an ordinary person - it is probably not a good idea to drive in the center of Moscow during rush hour right after getting a driving license, when you are still learning to drive. As you gain experience, you will be able to drive well even in heavy traffic. This is a task we will have to solve.
In terms of technology, it will take us about three or four years.
In May, it was revealed how and when the regulations concerning self-driving vehicles would be finalized in Russia. Yandex has also taken part in developing the plan for phased implementation of self-driving vehicles. During the conversation with ICT.Moscow, the representative of the company said that the document was being finalized and would be submitted to Vladimir Putin for approval at the end of August.
ICT.Moscow spoke with Yulia Shveyko, Head of Media Relations for Self-Driving Cars of Yandex.