🏹 ⛹🏻 🧔 8 questions about unmanned vehicles from Silicon Valley 👨‍💼 🤷🏽 ☝️

Recently he had a live broadcast with Mikhail Bakunin (Partnership Director of Nissan-Renault, who lives in Silicon Valley) on unmanned vehicles, which have long ceased to be science fiction and roam the streets of many cities around the world. A link to the broadcast at the end of the article, and for your convenience, highlighted 8 key questions about drones and wrote an expert’s answer to each of them several times.

Why should unmanned vehicles be electric?

Sensors, sensors and drone computers have a very high level of power consumption. So the car in any case should have an electric network. If a car with an internal combustion engine, then it will be necessary to build it, and not one, but two, since, according to safety requirements, all systems are duplicated in a robot. But the easiest way is to start creating a drone from a car that already has these systems and they need to be either finalized or slightly improved.

The second point is that today the most advanced technological solutions are installed in electric cars: electronic brakes, steering, gear shifting, etc. In other words, many of the technologies required by the drone are already installed and they can not be replaced, but modified. This greatly simplifies the work of engineers.

Is it realistic to create wireless charging for drones in the direction of travel?

The solution to this issue is associated with so many technological and infrastructural difficulties that in the next 5-7 years such solutions should not be expected. In the laboratory, nothing is impossible here. You can integrate charging into the road surface and the car will be charged while driving on it. But on a city-wide scale, implementing such an idea is incredibly difficult and expensive. Therefore, now they are mainly working to build more charging stations on the roads and make charging faster.

What are the levels of drones?

The Association of Automotive Engineers identifies 6 levels of car automation. Where 0 is a regular driven car, and levels 4 and 5 correspond to a car that does not need a driver or operator. At the same time, level 5 differs from 4 only in that such a robot does not need a high-resolution map.

(description of levels taken from www.bespilot.com )

In other words, a level 4 drone 99% understands where it is and what road conditions, signs and traffic lights are waiting for it out of sight. A car of level 5, in theory, can be sent to another unfamiliar city or country and he will figure out how to go.

Today, drones are at the 3rd level of autonomy, which, according to our guest, poses a certain danger on the roads, as drivers lose their vigilance. Level 5 is more of a marketing ploy than reality, and most companies are focused on creating a level 4 car, which should revolutionize the transportation industry around the world.

Are drones made of it?

Autonomous cars include two critical components: a vision system and software. What sensors work and why the drone needs them, we described in detail in this video:

As for the software, each drone has its own “brains” - a powerful computer (or rather two, because all systems must be duplicated), which it carries with it. All the calculations that a car makes on the road, it does not somewhere in the cloud or on a remote computer, but "inside itself." The “thought process” of a robotic auto involves collecting information from all sensors and comparing it. The machine not only needs to understand where it is and where it needs to go, it must understand whether there are any differences in the readings of its sensors and decide what to do with them. This is called localization of perception and this is the most difficult moment in creating software.

The second point is the calculation of the probabilities of events around the machine. Predicting the movement of each object around itself is a daunting task. Algorithms analyze everything: from blinking of a turn signal of a car ahead, to whether a child is standing alone on the road or with mom. After all, if he is with his mother, then he will not run across the road, and if he is alone, he can run. And the drone must foresee the possibility of such a development of the situation.

In other words, the car predicts a different development of situations, chooses the most likely and plans its actions in accordance with it.

But sensors and software are not everything. UAV developers also have to solve the problem of accurate response of all vehicle systems to the commands of the control computer. And it is not as simple as it seems. It is because of this that drones are tested for so long and in such a huge amount.

How do drones train?

Whether it will amuse you or scare you, engineers do not always know in advance what decision an autonomous car will take in a given situation. Machine learning is a complex field, it is fraught with many surprises even for those who have taught the machine to learn how to learn. Therefore, many situations are already dealt with by engineers after the fact - the machine makes a decision, and then people think about how and why it came to it.

Do drones need special roads?

No, the construction of special roads is such a complex and expensive issue that developers do not consider it. Their goal is to teach the car to ride on existing ones.

What niches does the development of drones open for startups?

Everything related to the infrastructure and maintenance of autonomous cars. From fleets that can charge, clean electric robots, to the development of optimal locations for serving and servicing drones (logistics, location of charging stations, etc.)

What problems do unmanned vehicle developers face?

Financial - the development and creation of drones are very expensive.
Regulatory - the lack of legislation governing the issue and operation of drones.
Social pressure - any accident involving a drone has been covered in the media for years, against hundreds of thousands of accidents involving people that no one is discussing.
Safety - so many are afraid of breaking into a robotic car. Here it is worth mentioning once again that all the calculations are carried out autonomously by the machine itself, it is simply impossible to take and connect through the clouds to the car in order to intercept control. The developers deliberately try to avoid this and in the serial autonomous car, all inputs and outputs will certainly be blocked so that it is impossible to get into the system. But the question still remains relevant and work on it.

Full live recording:

Thanks to Mikhail for an interesting interview, on his channel there are a lot of interesting videos about unmanned and electric cars: www.youtube.com/bakunin_live

8 questions about unmanned vehicles from Silicon Valley