Dialogue Mobileye Su Shuping: what is the most difficult part of autopilot landing?

category:Internet
 Dialogue Mobileye Su Shuping: what is the most difficult part of autopilot landing?


NetEase (NetEase news April 25th), recently, at Intels media communication conference, Su Shuping, Chinas general manager, accepted an interview with the NetEases intelligence and other media, explaining the disputes and views on the current different autopilot solutions, and the constraints on when the autopilot is on the road. The factor expressed her own point of view. What does the Mobileye change after being bought by Intel? Q: after Intel acquired Mobileye, Mobileyes technical team is also mainly in Israel. How do the two communicate in different places to ensure that the project is effectively promoted? Su Shuping: lets first introduce the company structure of Mobileye. Mobileye is divided into two product lines. The first is the product form of software + chip. The main customer group and end user are in fact a car factory, and a large batch of passenger car factories, such as general purpose, SAIC, BMW, Volvo, are all our customers. However, Mobileye and these customers do not produce direct sales contracts, because we provide only chip + algorithms, the car factory wants the final solution, the middle needs our Tier1 to do the integration, we have more than a dozen Tier1 partners around the world, such as Delphi and Magna, and we have three in China. A local Tier1 partner, such as Beijing Jingwei Hengrun. This is the first product line. The second product line, Mobileye also provides finished products, this product is mainly used in the rear loading. Because the cars life cycle is too long, normal is more than ten years. Now most of the cars that can be bought in the market are not equipped with ADAS, even some early warning functions are not available. How do we make these cars with the most basic ADAS functions to improve security? So we launched the after installed products, which is mainly operated by agents. Therefore, there are two product lines and different business models within Mobileye. Mobileyes R & D is indeed in Israels headquarters, and the R & D team will expand rapidly after our merger with Intel. But most of the offices in the world are sales oriented offices, plus local business support. Now, the products of the market are L0, L1 projects mainly, these technologies for Mobileye are ten years ago technology, the vast majority of the landing work is suitable for local models, this work is completed by the Tier1. Q: what is the help of Intel to Mobileye? Su Shuping: Mobileye is the most powerful technology, in other areas is relatively weak, like in the market, we are very low-key, do not do propaganda. But now with the help of Intel platform, there will be a good way to produce voice. In terms of business, Intels operation in China has a very strong resource in the industrial chain, which will certainly help Mobileye. How to use visual priority solutions Q: Mobileye proposed a VisionFirst autopilot solution at the beginning of this year. This solution is different from the multi sensor fusion solution that is currently used by the absolute majority of autopilot companies. Is there any advantage in the Mobileye scheme? Su Shuping: Mobileye and Intel will put forward visual priority because the eyes are our biggest information sensors, considering people driving. Intelligent use in the car is the car intelligence, this car intelligence will tell the car to make it have the ability to identify, decision, control, if it is to be treated like a person, the vision is essential. Since the company was set up for nearly 20 years, Mobileye has been making visual development to identify all the driving related objects on the road by simulating human eyes and brain processing, whether it is cars, people, lane lines, traffic lights, the edges of the road, and even the words on the road. Only in this way can we provide a comprehensive information assistance for our autopilot. The greatest advantages of vision are two points compared with other sensors. First, compared to radar sensors, the amount of information obtained is the most abundant; second, vision is the most suitable solution if it is from the cost of mass production. For these reasons, Mobileye will always adhere to the path of visual priority. However, what we should pay attention to here is that visual preference is not only visual, but it is hoped that people will not confuse the concept. Mobileye will also do redundancy, and will also do other sensors, such as radar and V2X (broadly speaking, the latter two are also sensors). Mobileye is also doing autopilot maps, but we are using REM (road network collection management) solutions to generate updates at lower costs and faster speeds through a crowd package. Q: if the cost of future lidar declines, will the technology direction of Mobileye change? Su Shuping: I insist on the solution of visual priority. I have been working in Mobileye for so many years. I am very sure of the companys technical route. Look at the problem from the users point of view. If it is a private car that is automatically driving, what kind of scene is to be faced, it needs to use the plan I said before. But if its an autopilot ferry, it can be built by means of a V2X+ camera and a geographical fence. Therefore, to consider a specific scenario to choose a solution, to design a good quality, price preference products, without the need to add laser radar. What is the biggest difficulty in automatic driving to the ground? Su Shuping: actually, the cost is very varied. In fact, we dont need to see the technology of L4 or L5. The technology of L0 can reduce the cost to a large extent. What is L0? Early warning is just FCW, LDW, these simple collision warning or lane departure warning, or collision warning of pedestrians. If a bus hits a person in the city, the cost of the claim will be very high. The biggest advantage of autopilot is to reduce traffic accidents. Our goal is zero death and zero casualties. Of course, it takes a long way. Taking the existing projects and overseas data, it is easy to reduce the accident rate by 50%. The cost of reducing the accident rate by 50% requires only a few thousand dollars after loading. But compared to a human life, this cost is much cheaper. The first problem of autopilot is the problem of safety. The second problem is the problem of efficiency. If automatic driving can reduce traffic congestion to a great extent, the cost savings can not be estimated. Q: in your opinion, what is the most difficult problem of autopilot landing? Su Shuping: the biggest problem of automatic driving to the ground is not the cost problem at present. The biggest problem is the problem of laws and regulations. There is also a very important issue of public trust. Q: autopilots landing is inseparable from the perfection of laws and regulations. What role do you think the government should play in this respect? Su Shuping: my personal view is that I hope the government is a conservative radical. The industry of automatic driving needs government encouragement, but it can not be radical. Too radical, once there is a little deviation in the process, it may make the whole industry retrogress for many years. This is something that practitioners do not want to see. Therefore, we hope that no matter what the regulations are promulgated and the definition of standards should be cautious, we should not be careful not to do so. Q: when it comes to trust, go to 4S shop to buy a car. Those salespeople dont know how SSC is opened. Why dont Mobileye promote it? Su Shuping: the education of the public and the education of consumers is a very long process. In the course of education, we have separated Tier2, Tier1, car factories and dealers between us and consumers, at least three layers. A technology after three levels of publicity will basically be out of shape, because there will be some adjustments in the middle tier for their own business in the promotion of caliber. Another possibility is that in the process of publicity, even if the material you gave is correct, the salesperson may be wrong when he conveys the information. For example, I experienced this case, the first time when AEB was introduced, we all knew that AEB was an automatic emergency brake, not an automatic brake, but when the consumer bought the car, it felt that the car was able to automatically brake and bought it. After that, the car did not step on the brake after that, and then rubbed it. After that, consumers complained about complaints from 4S and 4S stores, and complained to R & D all the time. When research and development colleagues were doing field research, they found it was wrong for users to use this product. If there is an automatic driving accident, how do we deal with it? Q: if there is an accident in automatic driving, how do we assign responsibility? Su Shuping: the judgment of responsibility distribution is very easy. There is a very clear responsibility interface between technology transfer and product transfer, which is very clear. And I believe that there will be a lot of technical means to judge in the process of design, because the data is there, and it is very easy to read the reason, such as the problem of recognition, the problem of decision-making, or the problem of motivation, which is very easy to judge. Q: before Uber, there was a fatal accident in an automatic driving accident. Can we avoid it if we use Mobileyes scheme? Su Shuping: the Uber fatal accident caused a great repercussion in the industry, and MobileyeCEOAmnonShashua wrote an article that explained his views on the event. Mobileye made an experiment that took a video of the police recorder and used Mobileyes current system to identify it. The result shows that our system recognizes the hit person, the car, one second before the collision. Therefore, with current technology, we can avoid tragedies or mitigate injuries. As for why Uber self driving car causes such an incident, it is still under investigation, and we look forward to seeing the result of the investigation. We know that when it comes to automatic driving, we will talk about ethical issues. In fact, we need to solve some practical problems in real life. Mobileye has proposed a RSS (responsibility sensitive security model). The most basic data are data and cases from all traffic casualties in the past of the United States Department of communications, classifying these cases and classifying the mathematical models after classification. Mobileye hopes that the self driving car will not take the initiative to cause traffic accidents. We have a mathematical model, but whether this model can be used in the whole industry, and to examine all industry companies and regulations in the industry chain to see whether they can accept such a set of ideas. How to build an autopilot ecology? Q: how does Mobileye build its own ecosystem in the auto driving industry chain? Will it compete directly with BAT and other Internet giants in building cars? Su Shuping: in fact, before the acquisition of Intel, the Mobileye ecology will be simpler, as I mentioned just just Tier2, through the model of Tier1 and the car factory. Of course, this is also based on the original technology, such as the level of L1 and L2, but after the level of L4 and L5, the requirements for solutions are different. After joining the big family of Intel, our product line is more complete. It turns out that we are a screw in the industrial chain, now the end to end of the entire industrial chain, we deliver more rich, more flexible, we can do more projects. As for competition, I dont think Mobileye is competing with BAT at the moment. Youll see companies like Intel and Baidu are partners, and theres no problem. I think competition comes from the future, and the biggest competition is the unknown. (end) pay attention to NetEase intelligent public address (smartman163), for you to interpret big events, new ideas and new applications in AI field. Source: NetEase intelligent editor: Guo Yin Shuang _NBJS5731 After joining the big family of Intel, our product line is more complete. It turns out that we are a screw in the industrial chain, now the end to end of the entire industrial chain, we deliver more rich, more flexible, we can do more projects. Pay attention to NetEase intelligent public address (smartman163), to interpret the big events, new ideas and new applications of large companies in AI field.