For the gift, start research and development of graphene batteries

Chapter 79 Without Big Data, Can Autonomous Driving Technology Be Impossible?
In this way, Jiang Ming settled down in Kyoto, concentrating on self-driving technology.

Hardware is one aspect, and the most important thing is software. The core issue is the strategic issue of software.

Although it is just the simplest shuttle bus, the team has been working on it for nearly a month, and the progress is still slow.

I don't know how many times the project meeting was held. At the meeting, the progress of the project was discussed.

"According to our plan, it is to use pure vision + lidar to generate models, and then use software to calibrate people and objects, so as to achieve the purpose of automatic driving."

Qin Ruya said so.

"This is also a relatively common design solution in the Xiaguo autonomous driving circle."

Qin Ruya continued to add.

"In terms of strategy, it is based on pure vision to quickly calibrate and identify the markings on the road. At the same time, the laser radar is used to further confirm the target."

A girl said that Jiang Ming also knew this girl, named Zhang Ru, who was a member of the autopilot team and the main programmer.

Therefore, Zhang Ru is very concerned about the logic in it.

"Why can't it be done at the same time? That is, pure visual perception and lidar are modeled and calibrated at the same time. After the results are consistent, the action is executed."

Qin Ruya asked.

Jiang Ming understood a little bit. According to Zhang Ru's plan, pure vision is the main force, and lidar is only auxiliary. The two functions are complementary.

As for Qin Ruya's plan, the two devices will be sensed separately, and then the system will judge and give the final action.

Similar to a competitive relationship, whoever recognizes the most wins.

"Director Qin, it's like this. The two methods can be used in parallel. In theory, there is no problem, but the requirements for the processing speed of the system are too high and cannot be handled."

Zhang Ru said.

"Moreover, the strategy is very complicated, and there may be many BUGs in concurrent operation, which is not worth the candle."

Zhang Ru continued to add.

Qin Ruya thought about it, and didn't continue talking, which was considered to have approved Zhang Ru's words.

In addition, in terms of big data, you still need to coordinate with Director Qin. We need enough servers to transmit data from the cloud.

In terms of real-time data, you need to communicate with customers. It is best to be able to connect to 5G signals, so that the data delay will be even lower.

Zhang Ru listed out her request item by item and asked Qin Ruya to resolve it.

Looking at the dozens of questions Zhang Ru raised, Jiang Ming couldn't help feeling that automatic driving technology is not as simple as he imagined.

Behind it is a series of technical support.

The guarantee of 5G signals, the deployment of remote servers, the construction of databases, and the development of management pages are all difficult problems before everyone.

Not to mention the cultivation of neural networks in autonomous driving technology requires the support of massive data. Without sufficient funds, it will be impossible to do it even if you want to.

This is still a few cars. If there are tens of thousands or hundreds of thousands of cars, the resources needed are massive.Just a large data center can drag down the average enterprise.

Therefore, although autonomous driving is the future development direction, ordinary companies really cannot afford it.

As for the hardware, Liu Ming also raised some questions.

The main problem is the sensor. The accuracy of the laser sensor is a bit poor and cannot meet the requirements.

In other words, it is the camera, which does not perform well at night and needs to be replaced.

It sounds simple to replace, but all kinds of verification work needs to be done again.There were many things, and Jiang Ming was busy with them.

"Hey, Mr. Zhang, I have a question for you."

Jiang Ming held Zhang Ru back and asked.

Zhang Ru is a girl with short hair. Although she looks weak, she is not sloppy at all when doing things.Obviously, it was influenced by Qin Ruya.

Knowing that Zhang Ru is the main player in automatic driving technology software, how could Jiang Ming let go of this opportunity and asked proactively.

"You said."

Zhang Ru's words were short and concise.

"Is there any special strategy for judging software?"

Jiang Ming expressed his doubts.

"Which aspect specifically?"

"For example, a traditional question is, when it is known that a traffic accident will inevitably occur, which target should the vehicle choose to hit?"

"Should we choose children or adults? Will the so-called old, weak, young and disabled have special privileges?"

Jiang Ming raised his own doubts.

Although this problem seems a bit ridiculous, it is something that autonomous driving technology must face.

This problem is the same as artificial intelligence. In the face of moral issues, whether to minimize losses or choose the optimal solution.These are two different results.

In the driving circle, there is a so-called saying of giving way to speed and not giving way.

In a traffic accident, in the face of inevitable impact, reduce the speed of the vehicle as much as possible on the premise of ensuring that you are legal.Instead of choosing to turn in an emergency, try to avoid the opponent's vehicle as much as possible.

The premise of this choice is that your driving is legal and the opposite is illegal.

In reality, this kind of problem is very common. It is obvious that the other party is going against the road or not following the traffic markings. In order to avoid the other party, you urgently evade, resulting in a violation.

This kind of violation caused this accident, so sorry, this kind of responsibility lies with you.

As a result, the person who violated the rules at the beginning patted his ass and left, leaving the well-intentioned party to deal with the traffic accident on the spot, apologize and lose money.Next year's insurance premiums will also rise, which is really thankless.

If there are casualties, maybe they will have to go to the detention center for a few days.

This kind of very realistic problem is ahead and has to be considered.

Jiang Ming wants to know how to deal with this problem with the thinking of a software engineer, and whether there is a corresponding mechanism.

It's just Zhang Ru's answer that made Jiang Ming a little confused:

"Why let the software make the judgment? The software can't judge this kind of thing."

"The strategy in the software, when faced with a possible collision, the first measure is to reduce the speed of the vehicle, and reduce the speed of the vehicle in the shortest possible time."

This point, Jiang Ming can understand, no matter what kind of traffic accident, if the speed is gone, the corresponding damage will be greatly reduced.

"The second point is to evade within a limited range without hindering the opponent."

This is also understandable. For example, if you have detected that the other party may hit you, and the sensor detects that there are no vehicles around, you can avoid this traffic accident by avoiding, then the self-driving vehicle will still avoid.

Of course, this kind of avoidance has many restrictions, such as speed, opponents and obstacles on the side, etc., which are judged comprehensively.

The premise of avoidance is that no secondary accidents will be caused.

"and then?"

When Zhang Ru said this, she stopped, and Jiang Ming asked.

"No, that's all."

Zhang Ru said.

"So, the software will only make judgments based on the conditions, and won't care about the specific target, and it's not even clear whether it can hit it or not?"

Jiang Ming summed it up, and then said.

"That's the truth."

Zhang Ru nodded in agreement.

Through the conversation with Zhang Ru, Jiang Ming generally understands the logic of autonomous driving technology.

When the conditions are met, actions will be taken naturally. As for the possible consequences of this action, there is no causal relationship with logic.

As for whether the old, weak, sick and disabled have higher authority, it is even more unreasonable.

"The execution of this kind of action depends solely on these external conditions, will there be any problems?"

Jiang Ming asked.

"Not only external conditions, but also many conditions for judgment. Conditions such as vehicle speed, road markings, weather conditions, distance between vehicles, etc. are all conditions for judgment. Only when all conditions are met, can the system be triggered operation."

Zhang Ru explained.

"Then how do you judge in environments with low visibility such as rainy days and foggy days?"

"Generally speaking, the automatic driving system will perform a self-test on the sensor, and only when the self-test meets the requirements can the automatic driving function be activated."

"Aiming at the use environment with relatively low visibility, generally speaking, we make special optimizations for this kind of working conditions, and basically the impact is not very big."

"But if it is true that the basic collected information cannot meet the requirements, then the automatic driving function will be stopped. At this time, it needs to be manually taken over."

"Of course, for working conditions in harsh environments, autonomous driving technology will take measures including slowing down and increasing sampling points to ensure safe driving."

"Let's put it this way, under severe working conditions, the sensors of the automatic driving system are more sensitive than human eyes. Correspondingly, the reaction speed is faster than manual driving."

Good guy, Jiang Ming understood what Zhang Ru said.

The meaning expressed by Zhang Ru is obvious. When the automatic driving cannot be turned on, it is even worse for people.

"Actually, manual driving should be more of a concern in low visibility conditions. Because, basically, manual driving collides with automatic driving."

Zhang Ru said leisurely.

Well, Jiang Ming was a little unconvinced, but in the end, he was delayed?

"No, I think the current automatic driving is very clumsy. Basically, it stays in automatic parking or high-speed automatic driving, which is still far from the level of automatic driving in downtown areas."

Jiang Ming said unconvinced.

This is indeed the case. Nowadays, autonomous driving technology should be called semi-autonomous driving technology. It means planning a route and letting the car come out by itself when the parking lot is parked automatically, or when there is no one around.

Other times, it's more of an auxiliary function.It's a pity to throw away and useless.Some people even claim that the current autonomous driving should be called mentally retarded driving.

There are mostly dazzling technical elements, but too few practical functions.

Although it is a bit exaggerated, it is enough to see the embarrassing situation of autonomous driving technology in the eyes of consumers.

"Hey, this is also impossible."

Zhang Ru sighed quietly.

"The environment for autonomous driving in Xia is not good, and many technologies are not easy to use. And how to put it, autonomous driving technology needs to be accumulated. It is not how much money you invest today, and it will produce results tomorrow. It requires continuous use of big data. Feed the algorithm to make the algorithm closer to practice."

"Let's put it this way, the program of autonomous driving technology is not very difficult, but if you want to train it and develop it into a system that meets the driver's habits, the data you need is not 01:30."

Speaking of this, Zhang Ru seems to be worried about the prospect of autonomous driving technology.

You must know that although these small scientific research institutes have certain resources, they are far from the level of mobilizing massive resources.

This is also doomed, they can only make small troubles, and it is too difficult to really fight against big companies.

Without data, it's hard for a smart woman to cook without rice.

“Does it mean that without big data, autonomous driving technology cannot be developed?”

Jiang Ming asked.

"It's not that it can't be done well, it can only be said that it is very difficult, so difficult that it is basically impossible to succeed. You must know that in the industry, algorithms are trained through big data."

Although Zhang Ru did not give a complete negation, according to his point of view, it is almost impossible.

"Then the smart shuttle project we are going to do this time, doesn't it also have big data?"

Jiang Ming asked curiously.

"Smart shuttles only operate within a limited range and do not require so much data. Of course, although they do not need so much data, it is also relative. In fact, we have already collected local road data. Using these The data is used for training, which is barely enough.”

Zhang Ru explained.

Oh, Jiang Ming understands, dare to love intelligent driving technology, not how good your programs and strategies are, the key is to have a good model and a large amount of data.

Jiang Ming has heard before that in the era of artificial intelligence, big data is valuable.Now it seems that this is indeed the truth.

But without big data, can't autonomous driving technology be developed?

Jiang Ming was skeptical about this.

Big data can train the behavior of the model, and can import past map data into the database for system query.

Although these data are detailed, how should I put it, after all, they are past data and can only be used as a reference.Really want the system to make judgments based entirely on these historical data, but it is also biased.

In fact, the automatic driving system is most based on the eyes of the vehicle, that is, the visual camera.

This is just like a human being, who perceives the outside world through the eyes, then the brain analyzes it, and finally makes a judgment.

Although the current autonomous driving technology has perfect visual sensing, how to put it, the establishment of the model is still much worse.

As a result, the automatic driving system relies too much on past experience, rather than making judgments based on model data.

The current situation is that the automatic driving system can often operate on clearly marked expressways and fast national roads, but it cannot be recognized when it comes to urban areas full of traffic.

An overly complex environment sets up obstacles for the establishment of models after visual sensing, making artificial intelligence not so intelligent.

For autonomous driving technology, Jiang Ming believes in better models and purer visual perception than using big data as a support.

Of course, this kind of model is very difficult. Jiang Ming hasn't figured out how to build this kind of model yet.

(End of this chapter)