Author: Chris Zheng
Prior to the 2021 Shanghai Auto Show, the video of the real test of the Beiqi Weichai Alpha S Huawei HI version equipped with Huawei’s automatic driving system ADS began to sweep the internet. The Huawei ADS exhibited extremely high algorithm robustness in the complicated traffic of the bustling city, like an experienced old driver.
The mass production of Huawei’s automatic driving system has attracted everyone’s attention and caused an upsurge no less than that of the Shanghai Auto Show two years ago. Huawei quietly visited and announced a comprehensive entry into the automotive industry.
This seems to be Huawei’s consistent style, not announcing anything until they make a big breakthrough.
On April 16th, Similarity Channel visited Huawei’s Shanghai Research Institute and interviewed the person in charge of Huawei ADS, Su Jing. Su Jing started from the HiSilicon chip and led the development of the Huawei Da Vinci AI chip architecture. Currently, he is the President of Huawei’s Intelligent Driving Product Line and Chief Architect.
Su Jing is the kind of interviewee that can quickly capture the media. He is quick-minded, speaks fast, has a very forward-looking vision for automatic driving, and is extremely confident. He is also not stingy about sharing his views on other companies in the industry. Of course, after speaking, he also very professionally adds: “This is my personal opinion and does not represent the company’s position.”
We have compiled the transcript of Similarity Channel’s and other media’s interviews with Su Jing, unedited. Enjoy it.
About the Alpha S Huawei HI version
Media: Which cities can our system support now?
Su Jing: Maybe I’ll briefly mention the composition of the system’s several modes first. It’s not a simple Robotaxi. It has the NCA, ICA, and ICA+ modes. I think the main experience you might ask about today is the ICA mode, which is completely automatic and similar to the Robotaxi experience.
When it goes into mass production at the end of this year, we will open up four cities: Beijing, Shanghai, Guangzhou, and Shenzhen. We will open up new cities approximately every three months. That’s the ICA experience. We also see that there are still many second-tier, third-tier, and fourth-tier cities in China where people still need to buy cars and use them. At this time, we will provide an ICA+ mode because our car will learn the entire traffic environment and construct a map on its own.
This car, as long as you or your neighbors have driven it, will automatically learn the situation of the road, construct a map in real time, and achieve an experience similar to a Robotaxi, but it will certainly be slightly worse because its data is not comprehensive enough. Especially for now, the popular high way, including Shanghai’s inner, middle, and outer rings, does not require a map and can achieve almost the same experience. So, this thing can be popularized across the country at the end of this year without any problem.
Media: Is it one city every three months after that?
Su Jing: Not just one city, every batch every three months.## Media: What is the approximate amount in a quarter?
Su Qing: It’s hard to say now, maybe around 6 at the beginning? I’m just giving an example, it’s approximately that level.
Media: What is the current testing range?
Su Qing: It has been generalized all over the country, in first- and second-tier cities.
Media: Can you talk about all the hardware configurations?
Su Qing: We have two configurations, the standard version has a computing power of 400 TOPS, and the luxury version has a computing power of 800 TOPS.
Media: Can you say something about the lifespan of the lidar?
Su Qing: The lidar can last up to 10 years on passenger cars.
Media: Is it in mass production?
Su Qing: It’s in mass production, available.
Media: When will it be delivered?
Su Qing: November and December of this year.
Media: In which regions of China will our vehicles be tested?
Su Qing: The first is Beijing, Shanghai, Guangzhou, and Shenzhen. These are the key areas, and we will also test the highways in all other major cities. This is the first batch to be covered. We will start running in second-tier cities in the second half of the year.
Media: When we talk about Beijing, Shanghai, Guangzhou, and Shenzhen, does that mean all roads in the city?
Su Qing: It means all roads in the city except for Beijing, which has specific regulations that do not allow entry within the fifth ring road.
Media: We have not experienced the AVP function these days. How is the R&D progress and mass production planning for this feature?
Su Qing: AVP is actually the first completed part. I thought everyone might be less interested in parking, but next time we can arrange for everyone to experience it. Because AVP will definitely be the best in our mass-produced cars, everyone should be impressed.
Media: Can it achieve L4 level AVP? Can people get out of the car?
Su Qing: I think so. People always like to talk about “hands-free”, “eyes-free”, “feet-free” in the industry. I know it’s good for financing or gimmicks, but to be honest, I’ve been working on autonomous driving for so many years and I really don’t like this term.
I think what needs to be solved is not to create a demo in a specific commercial area or building. That’s not what I want to do. What I want to solve is the commuting problem for everyone every day.
It’s obviously impossible to map out everyone’s workplace and home garage for each commuter, every office worker, every working-class person. No one can do that.
What I am willing to solve is the self-learning technology of the vehicle, to solve the automatic parking problem for each person’s office and home garage every day.The first step I pursue is not to have the person get off the car and leave the car, but to have the person arrive at the gate of the residential area. The car will tell you that you don’t need to take care of it anymore. Once you activate this feature, the car will automatically park in the parking space for you. This is the first problem I solved.
Media: Could you briefly explain NCA, ICA+, and ICA that you mentioned?
Su Qing: Simply put, the ICA mode is the one that everyone sees with pre-built high-precision maps inside the car. ICA+ does not have a high-precision map, but the car will automatically learn the map based on the environment where the car itself or other cars have driven through. This is ICA+.
In the place where you drive for the first time, there is always a place where nobody has been to, and no other cars have been there. This is the NCA mode. Tesla is now in NCA mode, which is divided into these three modes.
Media: What is the user experience like in the ICA+ mode?
Su Qing: You will find that ICA+ is a region between NCA and ICA. The more times your car has been driven or other cars have been driven through, the closer your experience will be to ICA. When you drive less frequently, the experience is similar to NCA, and it is a gradual self-learning process that improves over time.
Media: Can we understand that with or without a map, the confidence of the system may be different, and it may be easier to drop out in a certain situation?
Su Qing: Let me put it simply, if you go to a complex unfamiliar city and drive yourself, your speed will slow down and you will become more careful because you don’t know if there is a gap ahead or if pedestrians will suddenly pop out. The problem is the same for a car.
Media: Yesterday our engineer said that after downgrading to ICA+, it is impossible to achieve point-to-point driving?
Su Qing: That’s not entirely true. By point-to-point, it means that you can search for the destination in the map at any departure point. However, theoretically, there is no global map in ICA+, so it can only achieve this for places you have been to, such as the commuting points you go to every day, because you have driven through them.
But if you generalize it to all places, it is indeed impossible to achieve. You can understand its map as an incomplete map. This is easy to understand.
Media: Because the accuracy of the map is not as high as that of a high-precision map, does that mean that the ability may be slightly weaker?
Su Qing: The accuracy of the map is enough, but the data is incomplete. Let me give you an example. When you first drive, your own lane may be constructed, but the distant lane may have omissions, and the opposite lane may also have omissions. You have to drive more to accumulate more complete data.
It’s a bit like playing StarCraft in the past, do you remember? The map was black at first, but it turned white as you traveled through it. It’s just like that process.## Media:
Today the weather we experienced is pretty good. How about dealing with extreme weather like thunderstorms, typhoons, and also night mode and tunnel mode?
Su Qing:
Tunnel mode has no difficulty. There is no GPS in the tunnel, which is a location problem. But to be realistic, you cannot rely on GPS even when driving in the city or under the viaduct, except for demo purposes.
For extreme weather like thunderstorms, you can check out the video we released last year during the auto show. It is not a problem for us.
Night mode is also not difficult. There is no significant increase in difficulty when comparing night time with daytime.
As for collaboration with BAIC, it is not very clear yet since we work together to build this car.
If you have to divide it somehow, BAIC is in charge of the mechanical and chassis system, which is relatively traditional. Huawei helps to handle the computerized part of the car, including autonomous driving, the cabin, and the cloud on the back-end.
In the long term that’s probably how it will be divided, but it’s actually not that simple.
To be frank, differentiation is a serious problem. Do you think there is any differentiation in smartphones? The more complex electronic system, the more the development cost of each main body reaches billions of US dollars.
It’s not appropriate to differentiate at this point.
Huawei and BAIC’s collaboration is a series of car models instead of just one model. Investment is huge for both us and BAIC.
You will see many cars on the market in the first half of next year.
The completion rate of the car we tested is only 30% because of the algorithm. There’s no guarantee of 100% even for our own tested cars, with iteration updates every two to three months. The version of the car is consistent, but the adaptation period is only two months.## Media: What did the car companies teach you?
Su Qing: For example, at the beginning, everyone who was doing Robotaxi had a big tower of sensors on the roof, like a tower.
To be honest, we envied them a lot at the time because the algorithm would be much simpler. When we first started doing it many years ago, we hoped to put a tower on top, even if it was shorter, but it was strongly opposed by a major client, who absolutely did not allow us to do so.
So you see now that ADS cars look the same as regular cars, which is a very important lesson we learned from the car manufacturers.
About Huawei ADS Department planning
Media: How many people are in ADS? How long does it take for a car like this to be developed?
Su Qing: ARCFOXxing is the first car we’ve had in-depth cooperation with, and it’s likely been about three years since it was developed. We should be faster in the future, but there are always a lot of issues with the first generation. You can think that many of the rest of the imports will probably take about 24 months, and it will be difficult if it’s shorter.
Media: How large is the ADS team?
Su Qing: More than 2,000 people work on autonomous driving.
Media: Is the key research and development all in China?
Su Qing: Yes, it is.
Media: Can you talk about the division of people who make boxes, lidars, and algorithms in the 2,000-person team, and what percentage they account for?
Su Qing: You can think that there are about 1,200 people working purely on algorithms. Algorithms can be divided into several major blocks, which we call big perception, including vision and lidar, which fall under big perception. There is also a second prediction, and a third called PNC, which is regulations and control, which will be further divided.
You can think that the size of each team is approximately 200-300 people, and the remaining 1,000 people work on other things you just mentioned.
Media: Apart from the ARCFOXxing, will there be new car models and brands in the next phase?
Su Qing: Yes, there will be, and the company has already released three that I have seen, including ARCFOXxing, Changan, and Guangqi. There will be other major factories that you will see in the future.
Media: Is it our ADS going overseas or is it their domestic car model?
Su Qing: First of all, it’s their domestic car model.
Media: What is your opinion on Robotaxi? Will you consider doing operations?
Su Qing: First of all, let me express a personal stance: even if you hit me, I will never go to work on Robotaxi. Robotaxi is a result, not a business goal.
For Americans, the experience of taking a taxi is very poor. I’ve been traveling to the United States for many years, and the experience is very, very poor. In China, the experience of taking a taxi is very good, to be honest, and it’s not expensive. If you really talk about Robotaxi, China has already realized it, except that the “Robo” is a person, and there is no problem with this experience.# English Markdown
Today, you turned it into a computer. Frankly speaking, this experience did not improve at all. I firmly believe that this thing will not change the basic experience and basic foundation of travel in China.
Secondly, Robotaxi is the most difficult problem from the technological perspective because it needs to cover all Cornercases. That’s why I say it is a result, and you must take over before everything is perfect.
As time evolves and technology matures, it will be achieved one day. However, this time will be very, very long, and it requires a large number of cars, not just tens of thousands of cars that people are saying today.
You can only say that when you have hundreds of thousands or millions of cars and they have run for N years, and the data tells you that it is possible. Therefore, it is a result, and it should not be the goal.
In my opinion, all companies that have Robotaxi as a commercial target will be finished. The one who eventually achieves Robotaxi will be those who make passenger cars, and that market will definitely be mine, but not now.
Media: In addition to BAIC, Huawei also collaborates with several car manufacturers such as Changan and GAC to produce cars. Why did Huawei choose these car companies as partners, and what level of involvement does Huawei have in the collaboration?
Su Qing: There are many reasons for choosing customers. To be honest, the speed of domestic partners is indeed faster than that of international factories. This is also the reason why everyone can see that domestic car factories have their products released first, which may also be due to the speed of China.
The reasons for choosing a few car factories secondly are different. Like BAIC, BAIC is very sincere, and the cooperation is very good. Today, you can see that Huawei’s plan is not bad, but three years ago, what you saw might not be like this. Perhaps it was really a prototype at that time, and at that time, BAIC chose to trust Huawei and work closely with Huawei.
Moreover, they have really done a lot of work on the tuning and automatic driving chassis, which is a major reason for the in-depth cooperation with BAIC.
Changan is similar, and there are different reasons and business interests.
Media: What is the priority of Huawei’s autonomous driving in the Huawei car BU?
Su Qing: From my point of view, autonomous driving is absolutely the first priority, not a little bit of the first.
Media: What is Huawei’s future investment plan for autonomous driving?
Su Qing: We now have more than 2,000 people, and we spend about USD 1 billion a year. I guess the future growth rate will maintain around 30% per year.
Where does Huawei’s autonomous driving rank domestically?
Su Qing: It is definitely the first.
Media: What will be the future payment model? Will consumers pay once after buying the car, or will it be a subscription model?
Su Qing: There are two types. The first is the one-time payment model, and the second is the subscription model. Both will exist.# Media: Is the subscription model related to Huawei?
Su Qing: Yes, it is related to the car manufacturers’ revenue sharing. Why would my work not be related to it?
Media: When will the ADS department become profitable?
Su Qing: I am not in a hurry. As with everything Huawei does, it takes 10 years to be profitable. My only focus now is to make the technology world-class and solve real problems. Regarding autonomous driving, I don’t think profitability will be an issue.
For example, back in 2006 when Nokia was popular, we proposed making smart phones, but many people thought we were crazy. They consulted with a firm and reported that the user penetration rate was only 0.000 something and that it was just a toy for techies. They also asked, “what market is there for this?” First, you determine whether this is the right direction. If it is, then you don’t need to worry about the market; it depends on whether you can do it well.
Self-driving Regulations and Responsibilities
Media: How is liability for the safety of the vehicle assigned?
Su Qing: We always focus on improving the user experience and achieving the Level 4 autonomy, but from a legal perspective, it is currently Level 2 with no ambiguity.
If autonomous driving is to develop rapidly and bring a better experience to users, the functionalities, experience, and legal liability must be decoupled. Otherwise, car manufacturers will be extremely cautious and provide you with the safest, but virtually useless functionalities. Many car manufacturers are doing this today.
Media: When will autonomous driving move beyond Level 2, especially considering that regulations have become more relaxed?
Su Qing: To be frank, people who attribute the lack of progress to regulations are deceiving themselves; it is a technological problem.
As you can see, our vehicles have already achieved Level 4, but I can tell you that I still wouldn’t let the driver leave the car. Even if the driver can take control of the vehicle once every 1,000 kilometers, it will quickly become an issue. Until we achieve a very high level of MPI, there can be no talk of Level 4, only demos.
Whenever this issue is raised, people tend to cite regulations, but China’s laws are already very lenient regarding autonomous driving. The country is very supportive of autonomous driving. Bringing up laws isn’t relevant in this context.
About Huawei’s ADS Technology
Media: Are the self-learning maps uploaded to the cloud and then distributed to all vehicles?
Su Qing: It depends on the choices made by different car manufacturers. The maps can be kept at the vehicle end or be fused in the cloud and then distributed.
Media: Does Huawei provide the best solution, such that with just vehicle end or mass-produced cars, high-precision maps can be created without requiring dedicated mapping cars?
Su Qing: If you are only solving the daily commute, point-to-point route problem, then it is possible.## Media:
Is NCA collecting high-precision map data using our team for Huawei’s own applications?
Su Jing:
We have two parts to our system, let me briefly introduce our Roadcode system which consists of two parts: Roadcode HD and Roadcode RT.
HD can be understood as traditional high-precision maps created by a specialized map-making team offline. Roadcode RT is the car’s self-learning map. These two things are a combination of the two.
I haven’t been in this industry before and I didn’t realize that the infrastructure of the entire city changes so rapidly. I found that the roads of the entire Shanghai were constantly being renovated, and the speed of replacing traffic lights was much faster than I imagined. If you only use traditional technology like Roadcode HD, you will quickly fall behind.
Therefore, Roadcode RT itself will continuously learn and update HD, depositing the data, and going through this iterative process.
Media:
We found last night that the car is more hesitant when encountering delivery boys on the side. Is there any good solution?
Su Jing:
You are right. You will find that there is a blind spot in the laser coverage behind the production car, and it needs to be supplemented by vision. In addition, the car is not in its final production state. In fact, the actual tuning of the chassis of this car to be used for autonomous driving is after the Spring Festival, only two months away.
So, you can understand that the completion degree of the algorithm is only 30% – 40%. This problem will definitely be solved when you buy it.
Media:
Is the solution to add sensors?
Su Jing:
No, we optimize the algorithm.
Media:
Can accurate position detection of objects on the side be achieved only by relying on vision?
Su Jing:
In fact, this car has two circles of visual sensors, one for long-distance and one for fisheyes, which we also used.
You will find that the characteristic of vision is that the measurement error increases as the distance increases. When the distance shortens, the measurement accuracy will quickly improve, even surpassing that of laser.
This problem is a problem for the adjacent lane or near cut-in. In this case, there is no problem with short-range visual ranging. This is the principle.
Media:
So, during the driving condition, will we use the 360-degree view not only for parking?
Su Jing:
Of course, otherwise it would be a waste.
Media:
I want to ask about front fusion. Since it is front fusion, will all this information be summarized and digested into a set of computing centers or neural networks?
Su Jing:
You can think of all the information as input to the network, but it is not a single network. Different networks perform different functions. In fact, we have over 60 networks.Media: You mentioned that you can achieve unmanned driving for 1,000 kilometers. How was this data obtained?
Su Qing: To be honest, MPI is currently the best indicator for measuring autonomous driving, but there are many calculation techniques and methods involved in MPI. That’s why I am unwilling to discuss this issue.
You can see that MPI is related to several factors:
Firstly, it is related to statistical methods, and secondly, it is related to time and space. Time and space refer to what kind of road segment you choose to run at what time. All of these are related, and the value can differ by an order of magnitude.
Why is MPI related to statistical methods? We can see the statistical results from California. When you have hundreds of cars, you can select good samples and good time periods to statistically calculate MPI, and the MPI value will look very good.
When we calculate the MPI value internally, it is not meaningful to be frank. Instead, we accumulate historical data for all cars over all time periods. This statistical MPI value is the real MPI.
What I can tell you is that any MPI value in the autonomous driving team is a core secret. To be frank, I cannot tell you a specific number. It is not a simple number, but a large table. In the so-called California statistical method, I can indeed achieve 1,000 in Shanghai.
But in the real historical statistics, I can only say that I have not yet achieved 1,000. I must tell you this, and I dare to bet that no one in the world, including Waymo, can achieve 1,000.
Media: Is there a difference between your self-learning and Tesla’s shadow mode?
Su Qing: Honestly, Tesla’s mode so far has only been seen as a concept without detailed explanations. From our practical experience, there are at least a few things. It doesn’t matter if you call it shadow mode or car-side intelligence.
We have two major technologies. One is the Roadcode RT I just mentioned, which solves the self-learning and self-construction problems of the entire traffic static environment. AVP also relies on this to achieve.
Another one is what we call DDI, which may be more like the shadow mode you mentioned. DDI will continuously learn the driving behavior of the car owner. It may not be a takeover, but the behavior of the car itself may differ from the owner’s behavior. It will capture the owner’s behavior for iteration, which is perhaps what you call a shadow mode.
Media: Will you cooperate with other car manufacturers in the future, using the data from BAIC to create algorithms and push to Changan and GAC’s cars?Su Jing: We have a mode called Club, and car manufacturers can choose to join or not. If they join, everything is shared. If not, then you are on your own and others will not share with you.
Media: Is Club initiated by Huawei?
Su Jing: Aggregating data from car manufacturers is something only Huawei can do.
Media: Your forward perception is unique.
Su Jing: Where is it unique?
Media: Four cameras, telephoto + wide-angle + binocular. They are not commonly seen in mass produced cars.
Su Jing: Yes, because it is difficult to have binocular cameras, and most companies have not been able to achieve it. We just figured it out.
Media: What problems did binocular cameras solve for you?
Su Jing: Binocular cameras have a lot of problems. To put it simply, there are calibration problems from a mechanical perspective, and from an algorithmic perspective, using binocular cameras effectively is not easy because the inherent problem that binocular cameras have to solve is depth measurement, and measuring depth is actually a difficult problem to solve in terms of stability and generality. Most companies that use binocular cameras can only achieve a depth measurement of around 20-30 meters, whereas we have greatly surpassed this data.
Media: The current architecture has three lidars and many cameras. Will the sensor count and type be basically the same for other models?
Su Jing: It is almost the same. You will see that it is almost the same in this generation of cars. We usually make minor upgrades every 18 months and continue to iterate.
Media: You just said that autonomous driving may require millions of cars, and Tesla now has one million.
Su Jing: Good question. I remember someone saying something very reasonable before. What is big data? The focus of big data is not the word “big”, but the quality and comprehensiveness of data, which is the essence of big data, and autonomous driving is very similar. There are two key problems with the data. The first is the quality of the data itself. The second is the dimensionality of the data.
In these two issues, I think Tesla’s data has major problems.
What is dimensionality? Simply relying on a few visual data collection points, when high-precision positioning is not provided, the dimensionality of this data is very low. It is obviously seen that the data dimensionality of ADS is several orders of magnitude higher than that of Tesla. Data dimensionality is extremely important, as it represents the richness of information and degree of differentiation.
Second, data quality. You will find that the data itself is generated by algorithms, and the low complexity of the low-level system itself results in the low quality of the data. I guess Tesla’s data is already saturated and does not improve the system’s capabilities.
In fact, what we lack now is not data, but there are many difficult algorithms that need to be solved. I absolutely do not lack data.
Media: To solve these problems, the first logic is to have proper perception, and the second logic is to have correct predictions of other vehicles. Which one is more difficult?Su Jing: Good question. On the first day of work, we found it very difficult. After working for a while, we realized that predicting was difficult. After finishing the prediction, everyone found that regulation was also difficult. Now that we have completed regulation, we find that perception is also quite difficult, and we are constantly cycling through these challenges.
As for the overall complexity of the industry, in terms of theory and technological maturity, predicting and regulation are the real challenges, which many people are not aware of.
Media: The algorithms mainly rely on neural network deep learning. Sometimes deep learning can be like a black box. Do you think there will be breakthroughs in the algorithm in the future?
Su Jing: Firstly, automated driving systems are not only based on neural networks. The neural network is only one part of it. In terms of computing power, it accounts for the absolute majority, but in terms of code size, it is not. Let me clarify this point first.
Secondly, I firmly disagree with the statement that AI is a black box. Its computing model has changed from scalar calculation or linear calculation based on CPU in the past to probability calculation based on linear algebra. It is fully explainable from a probability perspective, with no problems at all.
You can think of it as something related to probability theory and statistics, and it is not that probability theory and statistics cannot explain it. I completely disagree with this view.
About Redundancy in Automated Driving Systems
Media: So, for long-distance perception, we rely on LiDAR?
Su Jing: No, we don’t look at it that way. Previously, people always asked a question: is your perception front fusion or back fusion, or what about redundancy technology?
First, I don’t think sensors have redundancy. That’s nonsense. Second, we abandoned back fusion technology two years ago. Now, we all use front fusion technology.
The characteristic of front fusion is to put all the information together and send it to the NN network for processing. It is not a simple problem of which sensor uses which information. You can also understand it as an Attention mechanism between sensors.
Another point is that the characteristics of different sensors are different. For example, millimeter waves are sensitive to speed but measure poorly, while vision measures semantics relatively well. LiDAR is good at geometric measurements, and it will fuse these together.
We directly use the raw data of millimeter waves, which is its original point cloud.
Media: Will it be difficult to obtain raw data from suppliers?
Su Jing: Two issues. First, most Tier 1 suppliers are not willing to open their raw data to you, but since Huawei is relatively large, they are willing to open it. Second, the raw data of millimeter waves is quite dirty and difficult to process. We are using NN to process it.
Media: Is binocular vision used for LiDAR redundancy?
（Note: This question is incomplete or ambiguous）# Translation in English
Su Jing: This thing can’t be called redundant. In fact, different sensors have different performances and advantages and disadvantages fluctuate.
Media: A leading self-driving company proposed true redundancy, making radar and LIDAR as a subsystem, and pure vision as another subsystem, independently testing both subsystems’ takeover rates, and multiplying them to achieve a statistically significant reduction in the required test mileage.
Su Jing: Frankly speaking, I guess that is what their marketing team wrote, not their R&D team. Otherwise, I will doubt their R&D capabilities. Indeed, what really determines your takeover rate is not just your perception system, but is closely related to your rules and controls, even more significant than the perception system itself.
These systems are not designed as true redundancy, are they?
Secondly, the vast majority of those difficult cases cannot be handled by adding 80 times the number of sensors. I bet with you. Therefore, the logic of using multiplication as an algorithm to do statistics is absurd.
True redundancy is a very marketing term. If you want to do perception well, you should do sensor fusion instead of redundancy, which is a serious waste of sensors. Their technological level is definitely not like this.
About ADS Competitors
Media: Some multinational companies are still discussing the concept of L3, while Huawei insists on continuity optimization. What do you think of this difference between traditional factories breaking through responsibility and we towards continuity?
Su Jing: In fact, European companies’ ideas are not entirely consistent. This is my personal evaluation and does not represent the company’s position.
In my opinion, among the three major companies in Europe, VW’s ideas are relatively advanced, which is related to their years of exploration in automatic driving. Other companies’ ideas are still in an evolutionary process.
In fact, Tesla, sorry, still needs to be mentioned. I think Tesla has taught everyone, including us and many other car factories, many things.
If you look at car factories, you will find what is fundamentally changing in the industry. Look ahead and it becomes clear.
To digress a bit, in the past, there were steam engines, then the energy revolution or power revolution, and then the computer was invented, and the computer changed everything. Actually, for the past thirty to forty years, this has been the process, the computer is changing everything. The last time they changed the phone, this time they changed the car. This is our view with Tesla, and this is our view.
The traditional car factory’s view is that its base is a car, and then there are some singular functions of the computer, so they try to embed the computer into the car. This is the traditional car factory’s view.
Our view is different. Our base is the computer, and the car is the peripheral controlled by the computer. This difference in fundamental perspectives leads to differences in all perspectives.Therefore, traditional car manufacturers tend to create many small boxes with the idea of adding one function with one box. However, our perspective is to build a computer, a big computer to handle everything, attach it to the car, and this is fundamentally different.
Media: So, is this the basic reason why Huawei does not make cars?
Su Qing: I think not making cars is a commercial choice, as the market is larger without making cars.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.