Thinking of unmanned vehicle technology

Abstract: This is the Chinese Academy of Engineering academician Li Deii last week in the theme of Baidu Unmanned seminar excerpts. Omitted some technical details of the description, in order to highlight the logical relationship to the content order adjustment, for the user's reading experience to use some more

This article is the Chinese Academy of Engineering academician Li Deii last week, Baidu unmanned seminar on the theme of the speech excerpts. Omitted some technical details of the description, in order to highlight the logical relationship to the content order adjustment, for the user's reading experience to use a number of more modern expression.

Li Deii is an automation and AI scientist, his keynote speech from the point of view of scientific research personnel on the unmanned vehicle technology thinking, put forward a dual-drive dual-control, complete vehicle informatization of the gradual technical route. He and Baidu IDL executive vice president, unmanned project leader Yukei at a point to reach a consensus, is not to do completely unmanned cars, people and smart car relationship should be "Knight and horse."

Automatic driving cannot rely on modification

The last time Yukei came to my office and said to me, "Our IT people don't make cars, I'm afraid we can't get unmanned." In many schools, unmanned cars are divided into two groups, one software group and one engineering team, which is a mess in the game. We used to do the same thing when we were working on a fund project.

Auto automation and intelligence, (Internet companies) always say to subvert who. Google and Baidu are generally said to subvert the auto industry. But what does the car industry think of us? Don't always feel good about yourself. Google made 200 unmanned cars without a steering wheel and was extremely proud. But it's nothing compared to the sophisticated modern car industry. A bit of a problem, the recall is hundreds of thousands of vehicles, internet companies have been?

My experience is that without the mass production talk about unmanned driving, installing sensors on cars for unmanned driving is still pediatrics. The automotive industry has achieved scale and refinement for more than 130 years and a reliable operation. The electronics of automobiles began to take 1993 years and have not yet entered the driving chain. (Some internet companies) want to make a modification to the traditional car to automate the drive, which is unrealistic. A military academy I worked with was sure to build my own car from the start, and I said it was right.

Sensor Imperfect principle

Sensors are not perfect, now add a variety of sensors on the car, add a variety of automatic driving scenes, looks colorful, but the use of the window is very small. And there is no perfect sensor combination, you say we buy more than 100 sensors are added to the car can it? Volume increased, maintenance increased, power increased how to do? Some unmanned driving scheme, car 200,000, sensor on 1.2 million.

Sensors are not ideal, sensors cannot replace the brain, and perception does not replace cognition. Human decision-making is based on the brain, not just the senses. Because no matter how good the sensor is, it is only limited perception. Perception is to understand the surrounding driving situation of various types of information, cognition is understood after planning and decision-making, and finally action. So three spaces, the first perceptual space, the second cognitive space, and the third is the physical space.

What should be done is to synthesize a driving brain through the overall cognition and control of the automobile, which not only integrates the senses but also incorporates the prior knowledge of the brain and the perception of time and space in driving. The machine decision is based on the cognition, the information of one way sensor can't control the driving link directly. Emphasis should be placed on the sensor's perceptual processing chip to achieve software hardware, the intelligent system to directly into the car can bus.

One planning and two times planning

Newton's time is small data big law, big data Age is small law. Information becomes more and more complete, and knowledge becomes more and more fragmented. To simulate a person's entire brain, I think it's hard for us to live. But the mathematical operation can be called the brain, as well as the Weiqi brain, and so on, from the human brain function stripped out, can complete a variety of driving awareness of the machine, can be called the driving brain. Driving the brain completely simulates the brain to complete low-level, tedious and long-lasting driving awareness, not distracted, never tired.

How to form the driving brain to refer to the human brain. Human functional memory is divided into short-term memory and long-term memory, short-term memory is perception to judgment, long-term memory is experience and knowledge. Therefore, the unmanned route planning is divided into one planning and two times planning. A plan is based on condition and road conditions. Many taxi drivers who refuse to hire at morning and evening are a plan. Second planning is real-time, is the local path to go, the latter part is often unable to achieve, and may slightly deviate from a plan. A large number of two of programs can also be used to store driving experience, such as recording a common trajectory at a junction, like a knee bounce reflex, as a planning supplement.

One planning is a static Cartesian coordinate system, and the two-time programming is a logarithmic coordinate system of motion. For all feelings of man, including sight, hearing, and touch, are proportional to the logarithm of the corresponding physical quantity, and this law is the law of psychological physics. We just keep on moving a plan from the static road map to the movement of the Cartesian map, which also put forward a high demand for the algorithm.

It's better to draw a two-dimensional code on the road

The road map stored in the driver's mind becomes the basis of the driver's travel task planning, the more familiar he is with the road, the clearer the memory, the more the plan is in place.

The success of the second planning can not be separated from the fine map, but the domestic fine road map or blank, which is to achieve the bottleneck of intelligent driving. Is the ring road in Beijing the same width on the map? I have asked several traffic secretaries, knowing that they do not know how wide the main road of their city is, and none of them can answer it. Centimeter-level map, how wide the lane, slope angle, the crosswalk where, there is no driving restrictions. Where is the transformation, where is the construction, must know.

Doing surveying and mapping will certainly say that this is too much work. My brother (deren) is studying surveying and mapping, of course I know. What we're going to do is just the opposite, putting things in the map of the road into the cognitive coordinates of the car. Imagine if a two-dimensional code is placed on the road marking, the two-dimensional code contains various information, semantic information, location information. No more accurate than yours? That would be a small burden. We are now the traffic signs, people are hard enough to read, you want sensors to read?

The best map is the road!

It's not subversion, it's improvement.

Smart cars to consider too many problems, economic, technical, legal, ethical, and the whole body. If smart cars are subversive to traditional cars, it's too difficult. Subversion and Revolution are escape, and the reformed are often successful. The car is not to say that we must be subversive to innovate, when we learn to drive the car liberation, turning the steering wheel requires strength.

Now the car network is the car companies do not let the car itself, add some equipment, listen to MAP3, the Internet. It is imperative to use it technology to improve energy and power systems to improve active safety. For example, fatigue driving reminders, background front, automatic shift, automatic parking, automatic cruise. There are many other scenes, such as the wet road and automatic driving. Now the new energy car, you can think about how to improve the efficiency of charging.

Do not rush to do without the steering wheel of unmanned, passenger is the owner of the wheel. Earlier in the day, it was manual driving, and then it turned into automatic driving, a cold switch between manual and automatic driving, which we felt was not good. Our idea is that the dual-drive intelligent car can bus, allowing manual driving while the automatic driving system quietly work.

Nissan recently announced the withdrawal of the steering wheel and various pedals of the automatic driving plan, instead of the introduction of more practical value of security technology. Tencent recently made a OBD 4G, to share the car OBD, the car's driving conditions, emissions are all known. These things are good for city planning and environmental protection.

I really agree with Yukei just now that the relationship between people and smart cars is knight and horse. Any horse is a two cognitive system, it knows the way to go, but know how to listen to the host. You just have to send a command and the road won't worry you. This is what I mean by double driving and double control.