Author: Su Qingtao

During the auto show, there were many big news related to autonomous driving, and many topics worth further discussion, but in order to avoid the overwhelming bombardment of information in the past week, “Nine Chapters of Intelligent Driving” chose to release content at off-peak times.

The main topics of this article are:

1. “Traditional IT thinking” VS “Internet thinking”;
2. After Huawei’s “sharp sword”, Robotaxi companies may be forced to adjust their strategies;
3. “Huawei Inside” becomes “Huawei Outside”, blurring the boundaries between To B and To C;
4. Huawei is not as “domineering” as outsiders imagine;
5. Advanced autonomous driving is making efforts to “liberate the eyes”;
6. Need both the ability of “hardware and software integration” and the attitude of “software and hardware decoupling”;
7. The two trends of computing power stacking and hardware reduction are parallel;
8. Is lidar really just “redundant”?
9. If the decision-making algorithm is not done well, adding more sensors is useless;
10. The market for low-line LiDAR may be eroded;
11. Map vendors must accelerate the collection of high-precision maps for urban roads;
12. “Small business” in big scenes and “big business” in small scenes;
13. Intelligent cars cannot simply be analogized to smartphones;
14. The value of companies that “lack halo” may need to be re-evaluated;
15. “High-end” may become Apple’s disadvantage.

Strictly speaking, this cannot be considered a serious “article”, but rather a “patchwork” of fragmented observations and thoughts. However, we believe that many of the phenomena and viewpoints mentioned in the article still have great value for discussion.

“Traditional IT thinking” VS “Internet thinking”

Huawei ADS leader Su Jing’s comments in a media interview were taken out of context by some pan-financial media, leading to doubts about Huawei’s “radical” approach; but in the eyes of some people who have had contact with many autonomous driving companies, Huawei is “not radical enough”.

After the release of MDC 810, a friend from a domestic Tier 1 company specializing in ADAS told the author on WeChat: “I used to be in the IT industry, and I switched to autonomous driving last year. I have a clear feeling that the autonomous driving industry tends to have an internet thinking approach, even before developing anything, they start to promote it, so most of the parameters are only on paper; in comparison, Huawei uses the traditional IT industry’s thinking approach, and only announces it publicly after developing more than 60%.”

After Huawei’s “sharp sword”, Robotaxi companies may be forced to adjust their strategiesTo be objective, in terms of user experience, the automatic driving capabilities presented by Huawei and BAIC BJEV’s Arcfox brand are comparable to many leading Robotaxi companies, and in some aspects, the latter may even perform better. However, only a few investors and To B automotive technology media journalists have experienced Robotaxi companies’ vehicles. To C media have not had the opportunity to experience them, making it impossible to make a direct comparison.

Previously, many Robotaxi companies were unwilling to give test rides to others. However, according to my colleague Professor Sun Li’s prediction, after Huawei shows its capabilities, Robotaxi companies’ attitude toward test rides will become more open.

However, Robotaxi companies’ shortcomings lie in their insufficient ability to integrate the supply chain, resulting in high prices for testing vehicles. This limitation makes it difficult for them to expand their fleet of cars rapidly and obtain large-scale data quickly, thereby struggling to overcome the limitations of the geographical fence. By comparison, Huawei can acquire data through mass-produced vehicles, and as long as they can sell 10,000 cars with ADS in collaboration with partners within one year, they will have a chance to surpass all Robotaxi companies, including Waymo.

Of course, Robotaxi companies will not sit idly by.

The dismissal of Waymo’s former CEO John Krafcik indicates a major setback in the direct approach to L4. Next, Waymo may enter the field through acquisition or investment and turn to a progressive route. Baidu focused on L4 and AVP in the previous years, neglecting L2, which now appears to be a mistake. However, by partnering with Geely to produce custom cars, Baidu has also begun to turn to a progressive route.

Originally, Didi was also directly engaged in L4 but, by joining forces with BYD to create custom vehicles and even personally manufacturing vehicles, Didi also has a great chance to turn to a progressive route, first installing L2 kits on mass-produced cars and then gradually iterating towards L4.

Among startups, Momenta has been taking a progressive route from the beginning. All four founders are from Waymo. The company initially intended to make an L4 Minibus, but recently launched an L3 bus. It has effectively turned to a progressive route. In the future, more companies that were originally dedicated to L4 may consider the progressive route.

Unlike the giants, whose progressive route can be realized by “making their cars,” startups usually need to work with car companies as suppliers (usually Tier 1) to enable their progressive route. However, how to deal with Tier 2 and Tier 3, ensure that essential components meet safety regulations and control costs are major challenges for these startups. In some sense, for these startups, it is even more challenging to do L2 than to do L4.Perhaps, for these startups, a better approach is to collaborate with established car companies, where the startup only provides the technology solutions and leaves the supply chain integration to the car companies; or collaborates with traditional Tier 1 companies that have supply chain integration capabilities, where the startup only earns money from technology solutions and leaves the integration of the supply chain to the latter.

“Huawei Inside” becomes “Huawei Outside”, blurring the boundary between To B and To C

In the second half of last year, a friend from a Robotaxi company proposed a viewpoint: for technology companies, the boundary between To B and To C is becoming increasingly blurred. For example, companies like Momenta are To C businesses in Robotaxi, but L2 businesses are To B; for example, TuSimple’s self-driving truck business, if supplied to Transsion Communications, is To B, but if the network is built directly to end customers, it could be both To B and To C.

After the battle for LIDAR mass production began, several LIDAR manufacturers even considered advertising on To C media at the right time to “educate” consumers about the advantages of installing LIDAR on their cars compared to those without it, and how our LIDAR is different from that of other manufacturers. After “my LIDAR is the coolest” occupies the To C user’s mind, the difficulty of acquiring To B customers is greatly reduced.

Before the Shanghai Auto Show, BAIC BJEV was supposed to release the new Alpha S, but it was Huawei, as the “party B”, that was in the spotlight. Li Xingyu, Vice President of Horizon Strategy Planning and Market Development, mentioned in his recent article “Ten Trends in the Era of Car Manufacturing 2.0” that the brand promotion of Alpha S by BHAE, “with Huawei as the supplier, is in the lead”, which is rare in the past, “and their end-to-end intelligent products are essentially 2C.”

In the words of investor Wang Yuquan, in this wave of PR, “Huawei Inside” has become “Huawei Outside” directly. Considering that Huawei will continue to help car companies sell cars in the future, “Huawei Outside” is not a joke.

In an interview with the media on April 19th, President of Huawei’s Intelligent Vehicle BU, Wang Jun, said: “Our smart auto component business is inherently a 2B business, but the products we offer, such as autonomous driving, intelligent cockpit, and thermal management, are closely related to the end user experience. Therefore, our business is both 2B and 2C.”Li Xingyu commented in the article, “At the end of last year, Huawei’s Smart Car Solution BU was merged into the Consumer Business Department, which does not mean that Huawei wants to make cars, but rather shows Huawei’s understanding of the nature of intelligent solutions: we must face the user directly, design products with users at the center, and create value. Technically speaking, it is necessary to develop end-to-end intelligent solutions from 2B to 2C due to the inevitable requirement of data closed loop.”

Li Xingyu also commented on the issue of whether Huawei should make cars: “This is more like a false proposition. When you have become Wintel, do you still care about making PC machines?”

Huawei is not as “domineering” as the public perceives

The usual explanation for why many car companies have not cooperated with Huawei is that “Huawei is too domineering.” In an article published on April 24th by “Yan Zhi Autonomous Driving,” the China Region Autonomous Driving Director of a Tier 1 company mentioned that “Huawei is too domineering” during an interview.

I also once thought that “tech giants like Huawei must be very domineering in front of their partners,” but from my recent contacts with Huawei and its partners, I found that Huawei is not actually domineering.

For example, although Huawei can provide a full-stack solution for autonomous driving, the MDC box is not bundled with lidars and mmWave radars. Customers can use Huawei’s MDC box with lidars and mmWave radars from other companies.

For example, although Huawei is very powerful, when facing weak customers, they are very humble and willing to do things that Nvidia “looks down upon” and serve these small customers well.

Compared with passenger cars and trucks, unmanned mining trucks are a “very small field,” but the MDC platform was first deployed in unmanned mining truck scenarios. The CTO of an unmanned driving mining truck company with less than 100 employees told me that they had approached Nvidia and Desay SV, but were rejected because “our demand is too small and they don’t care,” but when they found Huawei, the other party invested a lot of manpower in supporting them, even if Huawei had never heard of their company name before.

Over the past two years, Huawei has had more than 40 employees in total to support this small mining truck company. Huawei also acknowledges that the demand of this mining truck company is not large, but they have given us a lot of forward-thinking ideas.

In fact, in each specific scenario of autonomous driving, Huawei has invested the most resources in the “least noticeable” mining truck scenario.

When Huawei was doing communication business in the past, they often invested a lot of resources to build base stations for “poor, elderly, and remote areas” (including both Africa and rural America) that even international giants could not see the value of. This gene is also brought into their autonomous driving business.

Therefore, Huawei’s “strength” does not necessarily equate to “dominance.”Translate the following Markdown Chinese text into English Markdown text, in a professional manner, keeping HTML tags inside Markdown, and outputting result only.

As for those car companies who aspire to master the autonomous driving technology, are they really unwilling to cooperate with Huawei just because “Huawei is too tough”? If Huawei is “weak”, will these companies give up self-researching and adopt Huawei’s full-stack solution?

Regardless of whether they belong to the new force or the old force, many companies refuse to cooperate with Huawei because they fear that “Huawei will definitely make cars.” However, the author’s thinking logic is as follows:

Huawei has repeatedly bet on its commercial credibility to declare that it “won’t make cars” anymore. But car companies insist on pronouncing that “Huawei will definitely make cars,” referring to the precedent of “Huawei doesn’t make phones”, and dare not use its components. In this way, Huawei produces many parts that cannot be sold, which then forces it to “produce cars” to achieve “inner circulation.” Consequently, you would have a formidable competitor if you go down this path.

On the contrary, if car companies are willing to believe Huawei’s promise of “not making cars” and use Huawei’s components with confidence, Huawei would discover that the total profit from selling parts is “much higher than that of manufacturing cars.” Therefore, why should it make cars?

Advanced autonomous driving begins to strive for “Freeing both hands and eyes”

The L2 level autonomous driving under the SAE standard “neither frees hands nor eyes,” and theoretically, L3 is “capable of freeing eyes.” However, in practice, car companies are worried that once they allow drivers to “free their eyes,” they will not have time to intervene when they need to take over manually. Therefore, the so-called “L3 level autonomous driving” in this stage mostly emphasizes “only freeing hands, not eyes.” It essentially has no difference as compared with L2.

Last September, Lucid said in the release of its first production car Lucid Air that it plans to provide “L3 level autonomous driving that frees both hands and eyes” within three years. Compared to those so-called “L3s,” which “only free hands but not eyes,” this is obviously a bolder idea. If it can be achieved, it will also be a qualitative leap.

Recently, according to Garage 42, the D80+ and D130+ launched by DJI Automotive are also considered “L3 level autonomous driving, allowing drivers to take their hands and eyes off for a short rest during system operation.” Nevertheless, the article did not mention who should be responsible for the accident when a D80+ or D130+ equipped vehicle is in autonomous driving and the driver “takes their eyes off.”

In theory, only when car companies or autonomous driving solution providers promise to “take responsibility for your eyes off” to users can consumers “take their eyes off” with peace of mind. However, from a practical standpoint, neither car companies nor solution providers dare to take this responsibility lightly.

Moreover, in the future, DMS will basically become standard equipment for vehicles equipped with advanced autonomous driving systems. If manufacturers allow drivers to free their eyes, should DMS still be installed? If it is installed, how should its function be defined?### Need Both “Hardware and Software Integration” Ability and “Hardware Software Decoupling” Attitude

Huawei, originally a company that mainly focused on hardware, surprisingly has 1200 employees working on self-driving algorithms, accounting for 60% of its self-driving team. Based on Ascend 310 chips and MDC platform, Huawei has developed its own software such as operating systems and perception algorithms, as well as self-developed sensors such as LiDAR, sensors, and cameras to build its true hardware integration capability.

In terms of its capability system, DJI (including DJI Automotive and Livox) is similar to Huawei, as they both have the ability to self-develop self-driving computing platforms, LiDAR, millimeter-wave radar, and software algorithms.

Horizon Robotics is also worth noting in terms of hardware and software integration. Its founders and co-founders mainly have algorithmic backgrounds, but they have created a hardware-centric chip company. In actual business development, they have fully utilized their advantages in algorithms to help car companies develop algorithms, which can also provide hardware and software integration solutions.

AI computing has high requirements for the collaboration of hardware and software. Good software algorithms can make hardware perform to a greater extent, and vice versa. Therefore, compared with companies that only do hardware or software, companies with hardware and software integration capabilities will have stronger competitiveness.

For suppliers such as Huawei and DJI, hardware and software integration capabilities are the foundation of providing full-stack solutions.

From the perspective of car companies, full-stack solutions have unique advantages over combination solutions of hardware, software, and toolchains from different manufacturers, with “end-to-end performance optimization, clear and efficient responsibility interfaces, and platform smooth evolution”, which is very convenient to use. In addition, full-stack solutions can allow their cars to achieve self-driving capabilities earlier, which is not something all car companies have the courage to do even if they invest a lot of resources in self-developing their own self-driving algorithms.

For these car companies, full-stack solutions are actually the best choice — being the “outsourcing factories” of tech companies is better than continuously declining sales or even persistent losses and bankruptcy.

According to media reports, Horizon Robotics founder Yu Kai and Innovusion founder Bao Junwei respectively proposed chip and LiDAR designs during their employment at Baidu, but their proposals were not adopted at the time. Since then, Baidu has invested in Velodyne and Hesai in the field of LiDAR and has cooperated with one of them to develop forward-looking production LiDAR. Next, Baidu might also design its own self-driving chips.The companies such as Huawei, DJI, and Horizon Robotics have the capability of integrating hardware and software, but they adopt an open business model with hardware and software decoupling. For example, Huawei and DJI’s computing platforms can install laser radar and algorithm provided by other companies, while Horizon Robotics has adopted an “unbundling” strategy to attract many customers from Mobileye.
However, Huawei also emphasizes that only those car models equipped with Huawei’s full-stack autonomous driving solution can carry the “HI” (Huawei Inside) logo. This means that even though Huawei has provided a solution for hardware and software decoupling, car manufacturers with weaker product strength that hope to enhance their brand value using “HI” are more inclined to use its full-stack solution when cooperating with Huawei.
In summary, the ability system of “hardware and software integration” must be maintained to ensure competitive technology and products, while the attitude of “hardware and software decoupling” is also necessary as the basis to gain the trust of partners.

In January of this year, NIO and SAIC ZhiJi launched the prelude of “auto-driving computing armament race”, however, whether viewed from the perspective of cost or power consumption, having too many chips may just be a transitional solution. When it is time for large-scale mass production, subtraction may still be required.

According to a Huawei employee quoted by the financial magazine on April 18th in an article called “The First Truth of Huawei’s Car Making”, “while ensuring the overall functional computing power, Huawei has already been trying to replace computing modules such as memory chips and chip modules with lower-cost modules, which is also for the consideration of car companies.” Meanwhile, according to a DJI engineer, “with DJI’s excellent algorithm, the calculation power requirements won’t exceed hundreds of TOPS even to achieve L3+ system”. This means that in the future, some companies with strong accumulation of software algorithm capabilities will not be passively involved in the “computing armament race”.

In previous years, the common saying for sensor schemes in the auto-driving industry was that “Laser radar is the main sensor, and cameras are auxiliary sensors”. However, in the past two years, more and more people have begun to say “using cameras as the main sensors and laser radar as redundant sensors”. However, the “redundancy” concept is being questioned. In late November 2019, Freetech CEO Zhang Lin explained to the reporter about the “main sensor/auxiliary sensor” term, “In the era of domain controllers, it is all about fusion perception algorithms. We won’t discuss who is the main sensor and who is the auxiliary sensor.” In July last year, when talking about this topic with the CEO of a laser radar company, the reporter was asked, “How to define who is the ‘main sensor’?”At the time, the author quoted an explanation from the head of autonomous driving at a certain new energy vehicle company: “When the recognition results of two sensors are inconsistent, the decision-making system decides whose opinion to follow, and that sensor is the main sensor.”

In response, the CEO of a LiDAR company said: “If the camera’s recognition results are followed when they conflict with those of a LiDAR, why not just remove the LiDAR? And if the LiDAR’s results are followed when they conflict, then calling the camera the “main sensor” becomes meaningless – it’s like saying ‘I make the main decisions in our home, but when there’s a disagreement, my wife has the final say’.”

This logic is quite convincing.

Recently, Su Qing, head of Huawei ADS, stated in a media interview: “I don’t think there is such a thing as ‘Redundancy’ for sensors, that’s nonsense. We abandoned fusion technology two years ago, and now we use only front fusion technology.” The implication is that if perception uses a front fusion solution, the concept of a main sensor versus redundant sensor becomes meaningless.

Recently, Bai Yang, author of “Garage 42,” also criticized the saying “LiDAR is a redundant sensor” in an article, stating: “The meaning of ‘redundancy’ is that you can still manage without it, but you have an alternative when you can’t manage; at this stage, not only is LiDAR not redundant, it has become the mainstay for implementing some scenes and functions, which is clearly ‘taking on the work of the regular but pretending to be the backup’.”

Usually, when autonomous driving companies talk about “LiDAR as redundancy,” they are implying that “our visual algorithm is already very powerful”, but this may be digging a hole for themselves. An algorithm engineer for a certain new automaker said: “Musk’s first principle is okay if you look at it from a perspective 10 years from now, but the problem is, if there is no revolutionary breakthrough in vision algorithms now, his technical path will be beaten by the perception fusion path.”

Next, companies will continue to work to improve their visual algorithm capabilities, but as long as they haven’t completely eliminated LiDAR, who is the main sensor and who is the redundant sensor will be more like a word game.

No matter how many sensors you pile up, if the decision-making algorithm is not good enough, it is useless

During a media interview on April 18th, Su Qing responded to the media’s mention of Mobileye “treat[ing] radar, LIDAR as a subsystem, and pure vision as a subsystem, [and] independently test[ing]” by saying: “What really determines the takeover rate is not just the perception system. It is closely related to regulations and controls, even to the point where the weight of regulations and controls is greater than that of the perception system.” “In the vast majority of difficult cases, adding 80 more sensors won’t solve the problem.”

Relying on stacking sensors cannot fundamentally solve the problem. I completely agree with Su Qing’s perspective.Last August, when I talked with Dr. Guo Jishun, from GAC R&D Center, about which is more difficult, perception algorithm or decision-making algorithm, Dr. Guo Jishun answered: “Perception is something easy to standardize, while decision-making algorithm is easy to make differentiation. Evaluating a person’s capability is not about whether his eyesight is good or not, but about whether his thinking is impressive or not.”

During L2 stage, as human is the subject responsible for driving, the development of autonomous driving algorithms of all companies focused on perception algorithm. Therefore, the one with better perception is considered more competent. However, as autonomous driving advances to the higher stage, the responsibility gradually transfers to the system, and the real challenges are in behavior prediction, path planning, and control, which refer to decision-making and control algorithms. In short, even if one has 20 myopic eyeglasses, they cannot marginalize the shortcomings of lack of intelligence and experience.

A certain new force in car-making said, “Those managers who do not understand the whole technology chain would think that perception is a limit. Of course, perception is also problematic, but it is not the main reason. We are so diligent in adding radars, cameras, microphones, speakers, error reporting logic, and AI brain to the car to make the perception of the car as close to the minimum perception ability of a person as possible. However, from a human’s perspective, would infinite pursuit of accurate perception data lead to algorithm traps?”

The low-line lidar market is expected to be eroded.

Recently, the concept of 4D millimeter-wave radar has become popular, and whether 4D millimeter-wave radar can replace lidar with low lines, such as 32 or 16, has also become a heated topic. The author’s opinion is that in certain scenarios, 4D millimeter-wave radar can replace lidars with low lines, but it cannot replace those with high lines.

Huawei’s Car BU President, Wang Jun, said in a media interview, “When we were developing 4D millimeter-wave radar, we demanded that its performance must be able to replace lidars with low lines, thus compelling lidars to move towards high lines, forming a virtuous cycle in which both technologies push for better performance, leading to cost reduction.”

As the market penetration rate of 4D millimeter-wave radar continues to increase, the market for lidars with low lines will shrink. Of course, lidars originally were being developed from low lines to high lines. Even if there were no 4D millimeter-wave radar, the halo of lidars with low lines would be gradually dimmed by those with high lines. This belongs to the lidar manufacturers’ “self-revolution”.

In the future, the lidar manufacturers will increasingly direct their focus towards high-line products, which will be a boon for the technological advancement and cost reduction of high-line lidars.

Mapmakers must accelerate the collection of high-precision maps of city roads.Currently, when major map vendors in China talk about how much of the road their high-precision maps cover, their data mainly focuses on highways and few mention urban road data. The reason for this is because previously, it was believed that high-level autonomous driving capabilities would first land on highways, so focus was largely put on highways. However, companies are now finding that the situation has changed.

Huawei’s autonomous driving technology did not solely focus on highways from the beginning, but instead mainly addressed the commuting issue for city residents. Americans spend more of their time driving on highways, so Tesla’s autonomous driving capabilities started with highway scenes. In contrast, Chinese drivers spend most of their driving time on urban roads, so Huawei’s choice of technology landing scenes is more suited to China’s reality.

In 2019, Huawei entered into partnerships with several domestic high-precision map manufacturers, but at the time, these manufacturers did not collect much urban road map data. As a result, Huawei applied for a high precision map collection qualification and decided to collect high-precision maps for urban roads themselves.

Next, as vehicles equipped with Huawei’s high-precision maps gradually hit the road and the range of high-precision maps expands rapidly, updating high-precision maps in a timely manner will be essential. If other map vendors’ high precision maps cannot match the coverage of urban roads, then they will be in a very difficult position.

Furthermore, if there are enough vehicles to scale up, high-precision maps can be automatically updated. The vehicle will identify significant differences between actual road conditions and the map, and then trigger collection vehicles to recollect data, which will greatly improve efficiency. However, if the number of vehicles is too small, the difficulty of “automatic discovery” will be high, and manual planning will be required, such as updating once a month–which may require re-scanning all roads indiscriminately, which will waste resources and energy.

It can be said that if the map vendors cannot provide their own autonomous driving overall solutions or have deep cooperation with auto enterprises, they will be very passive regarding updating high-precision maps.

In addition, during a car show, I met the CTO of a driverless mining company who told me an interesting piece of information. “End-to-end logistics is an approximately 100 billion RMB level market, but the market is spread out with strong regional characteristics. If you do well in one city, you may not necessarily do well in another city. Furthermore, the order volume for each community is very small, and operating costs are difficult to allocate. In contrast, the market size for open-pit coal mine earthmoving transportation is only 30 billion RMB, but each mine is very large. If you can take an order from one mine, you can earn several hundred million RMB in one year.”

This information is valuable. When autonomous driving companies are searching for application scenarios, they shouldn’t just focus on “how big is this market”, but instead evaluate how difficult it will be to secure a certain market share. In summary: if the chosen direction does not match their own ability system, they may only make small deals in a big scene, but if the direction is the right one, they may make a big deal in a small scene.### Cannot Simply Compare Smart Cars to Smartphones

In recent years, many industry insiders, especially influential media figures, have been comparing the future of the smart car industry to that of the smartphone industry. However, this comparison is not entirely reliable.

On April 18th, a journalist from “Late Post” asked Wang Jun, President of Huawei’s Car Business Unit, “When Huawei transitioned from the smartphone industry to the automotive industry, did you think there would be experiences that could be replicated, but actually found that the differences were significant?”

Wang Jun’s response was, “The smartphone industry is focused on the consumer market, whereas the automotive industry deals with industrial standards and regulations. When we entered the automotive industry, we found that the requirements were completely different. Besides the threat of exploding batteries, smartphones have a relatively small impact on personal safety, while many aspects of smart cars are closely related to human lives.”

Wang Qingwen, General Manager of the Intelligent Cockpit Product Division of Huawei’s Smart Car Solution Business Unit, added, “From an end-user’s perspective, smart cars are similar to smartphones, but from an engineer’s and product developer’s perspective, they are two completely different things. This is because their lifecycles are different, and having different lifecycles leads to completely different software development, software management, and hardware management requirements. Smartphones may only need to be upgraded once every two years, so the software version doesn’t need to take hardware into account. However, the lifecycle of a car is usually more than ten years, and if the software is updated, every version in the car needs to match with the existing hardware. This is a huge workload.”

Wang Jun said, “We are still exploring how to manage the software lifecycle, especially in the cockpit field, which involves frequent upgrades. Lifecycle management is very complicated, and there are too many versions to maintain. Therefore, defining software-defined cars and OTA is not a simple matter. It requires a lot of manpower, material resources, and processes to ensure that it can be truly implemented.”

The Value of Companies Without a “Halo Effect” May Need to Be Re-evaluated

In the past, when autonomous driving was still in the testing stage, everyone was competing on technology capabilities. However, at the stage of mass production, the importance of engineering capabilities has been highlighted. In short, technology companies with backgrounds in the internet and AI may have stronger technological capabilities, but traditional Tier 1 and automotive companies may have a “crushing” advantage when it comes to engineering capabilities.

A few months ago, a friend who was the CEO and CTO of a domain controller company from a German Tier 1 company said, “Our boss is very frustrated. We are already profitable, so why is our valuation not as high as some L4 autonomous driving start-ups that have been losing money, or even facing life and death?” I joked, “Because in the eyes of investors, your company lacks the “technological feel” and has no “halo effect”.”Compared with many founders and even CTOs of self-driving companies from internet backgrounds who are usually good at storytelling and even possess a “celebrity” attribute, many founders from traditional Tier 1 backgrounds are often “honest and reliable engineers” who are not good at speaking. In extreme cases, the investment and marketing departments of their companies may also be technical men who are “at a loss” when they meet investors and media, which may result in the undervaluation of these companies.

However, when self-driving technology shifts from the storytelling stage to the engineering capability stage, the value of these companies may be further tapped, and their valuation may even rise significantly.

“High-end” may become Apple’s disadvantage

The release of XPeng P5 and the cooperation between DJI and GM Wuling indicates that high-level autonomous driving is beginning to “popularize” in terms of price, which is of great significance.

The earliest mass-produced cars equipped with autonomous driving systems, such as Model X, Model S, and later Audi A8, all cost more than RMB 800,000. Cadillac CT6 costs more than RMB 400,000; XPeng P7 costs less than RMB 350,000; XPeng P5, equipped with lidar, is expected to be priced at around RMB 200,000. Next, the mass-produced car jointly developed by DJI and GM Wuling may be priced at around RMB 150,000 or even lower, and Xiaomi’s first car is also expected to be priced at this level.

Previously, when the prices of chips and sensors were relatively expensive, autonomous driving systems could only be deployed on high-end models whose users were not sensitive to prices. However, with the intensification of competition and the decrease in supply chain costs, more and more mid-to-low-end models can “afford” autonomous driving chips and lidars.

He XPeng once mentioned that “the higher the customer’s level, the more interested they are in autonomous driving. The higher the super high-end customers, the absolute demand for luxury and power feeling of the car brand.” In the early stage of the industry, this viewpoint is basically correct, but we cannot understand it as “mid-to-low-end users are not interested in autonomous driving at all.” They just don’t have enough money to pay for their interests, so they have to “pretend not to be interested.”

As the cost of autonomous driving kits rapidly decreases, mid-to-low-end users who were previously “not interested” in adopting autonomous driving will soon become “interested.”

Unlike smartphones, the ability of autonomous driving systems largely depends on the size of the car fleet. Therefore, from the perspective of car companies, under the premise of equal hardware configuration and similar user installation rates, those models with higher sales will make greater contributions to the advancement of their own autonomous driving algorithms.

After all, the competition in autonomous driving capabilities among various companies is short-term, which depends on algorithm talent and hardware configuration, while in the long-term, it depends on data volume, which is determined by the size of the car fleet equipped with autonomous driving kits.Therefore, if carmakers believe that improving autonomous driving algorithms is their top priority in situations where resources are limited, they should prioritize equipping best-selling models that are likely to become popular with autonomous driving systems rather than equipping them on the “top-end” models. The value of accumulating data from a few thousand autonomous vehicles that are high-end but sell relatively poorly will be limited. If it is possible to make autonomous driving a standard feature in all vehicles as per the ideal plan, then that would be the best case scenario.

From this perspective, although Apple has a wealth of talented employees, their autonomous driving capabilities may not necessarily enter the top-tier. This is because Apple’s style dictates that their products must be “super high-end”. Suppose that Apple includes autonomous driving suites as a standard feature on a RMB 600,000 car, while DJI or Xiaomi includes an equivalent or slightly inferior autonomous driving suite on a RMB 150,000 car. Who would have the advantage of obtaining the scale of data and whose algorithm would improve more quickly?

Of course, DJI or Xiaomi would also consider how to deal with the T-Box in the vehicle, how to transmit data, and how to process the received data – without processing capabilities, having significant amounts of data would serve no purpose. In addition, the first vehicle equipped with an autonomous driving system should have a centralized architecture to support whole vehicle OTA updates later.

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.