Interface News Reporter | Lin Teng, Zhang Xilong
Interface News Editor | Xie Lingning
Guangzhou, Nansha, Shenzhen, Huizhou, etc., distributed on the east and west banks of the Pearl River estuary, the seemingly ordinary highways, as well as their parks, will become the key landmarks of intelligent driving in China and the world in the future? At least for the entrepreneurs in the middle of it, they are willing to believe in such assumptions.
Among them, Nansha District of Guangzhou is located on the west bank. It is the youngest urban area in Guangzhou, and it was first famous for its seafood, but in recent years, the most talked about are Lexus taxis that can drive on the road without a driver.
Once a status symbol and known as "Lexus", Japan's high-end Toyota brand has been transformed into a "Robotaxi" by a group of Chinese engineers, a luxury driverless taxi that locals can ride at a low price.
The keywords behind this have become more and more familiar to people who care about autonomous driving: returning to Silicon Valley and developing self-developed algorithms. In a Mediterranean-style office building in Nansha, Peng Jun has led his Pony.ai to lie dormant for seven years, and the cumulative journey of autonomous driving has reached 35 million kilometers.
Even established brands such as Toyota are beginning to admit that Pony.ai's autonomous driving technology has a leading edge. Just recently, Toyota and Pony.ai formed a joint venture to launch thousands of Robotaxi across the country. Today, Pony.ai is one of the youngest unicorns in China, with a valuation of $8.5 billion.
Cross the Pearl River along Nansha and arrive at Nanshan District, Shenzhen. In an industrial park, Qiu Chunchao, a Chaoshan man, was tinkering with kung fu tea in his office, speaking a mouthful of Chaoshan Mandarin, and excitedly showed reporters a lidar that only costs $200 and is only the diameter of a one-yuan coin, which is one of the core devices of autonomous driving.
Qiu Chunchao was a businessman in Huaqiangbei, and his elder brother Qiu Chunxin graduated from Harbin Institute of Technology (Shenzhen) with a doctorate, and the two brothers spent ten years developing lidar, the price from nearly 100,000 US dollars in American companies to only a few hundred dollars, and used self-developed chips to promote lidar from high-tech industrial applications to mass production of vehicle front-loading applications.
Since 2021, Suteng Juchuang products have been purchased in large quantities by Geely Group, BYD Group, SAIC Group, new power brands Xiaopeng, Lucid and other car companies for intelligent driving, becoming the global sales champion of on-board lidar. In 2024, Suteng Juchuang will be listed on the Hong Kong Stock Exchange, with a market value of up to HK $35 billion.
If you cross Nansha and follow Shenzhen all the way to the east, you will reach Huizhou in an hour's drive. Singaporean Chen Chunlin's 30-year-old production line for Germany's Audi to make supporting audio systems in China has been transformed into a production line that produces intelligent cockpit controllers that intelligent driving companies are vying to order.
Desay SV, the company helmed by Chen Chunlin, has now become the core partner of NVIDIA, the top company in the field of autonomous driving chips, in the field of intelligent cockpits.
Since 2020, the intelligent driving industry has begun to surge, and Desay SV's earning rate has increased by 30% per year, and its market value has skyrocketed 10 times to 70 billion yuan in three years. Driven by this company, Huizhou's manufacturing industry, which was originally unknown, has suddenly become a high-end player in the era of intelligent driving.
If you look at the technological developments of these three companies together, you may be able to believe in the fact that the "singularity moment" of autonomous driving is coming.
"These companies are already one of the global vanes of the autonomous driving industry," Zhu Xiaorui, president of Singapore's GAIR Research Institute and former professor of Harbin Institute of Technology (Shenzhen), told Jian News. According to McKinsey, by 2030, China's new car sales and mobility services related to autonomous driving will generate more than 500 billion US dollars.
One of the facts on which these judgments are based is that Guangdong's local autonomous driving industry has been in the industry for decades. From the east bank to the west bank of the Pearl River, a complete industrial chain of algorithms, sensors, and intelligent cockpits has been formed. Whether it is local car companies such as BYD, GAC, Xpeng, and Huawei Wenjie, as well as top foreign companies such as Volkswagen and Toyota, they are continuously purchasing a large number of products in this industrial chain to realize the intelligent driving function of vehicles.
Flooding into this industrial wave are not only returnees from Silicon Valley, but also local scientists, and even "masters" from traditional manufacturing.
An industrial chain that has changed the way of driving for a century is accelerating its rise.
"Autonomous driving must be in China, where there is the world's most abundant and complex road data, and research and development can be fed back through testing." Peng Jun said to Jiemian News. As for the technical route chosen by Pony.ai, he had a clear idea from the day the company was founded.
Peng Jun believes that the number of cars per capita in Chinese is still relatively low, but the willingness to travel is very active, so the demand for shared travel is great, so in the shared travel scenarios such as buses and taxis, the demand for autonomous driving is higher.
Guangdong has abundant resources in the automotive industry chain. Around 2000, Japan's Toyota and Honda both set up joint ventures in Guangdong, after which BYD also used its battery advantage to cut into the field of new energy vehicles, and in the past few years, it has also given birth to new forces such as Xpeng.
Before founding Pony.ai, after graduating from Tsinghua University's Department of Civil Engineering in 1996, Peng Jun went to the United States to pursue a doctorate after graduating from Tsinghua University's Department of Civil Engineering in 1996, and got his first job at Google in 2005. In 2012, Baidu set up a research institute in Silicon Valley, and Peng Jun joined as one of its earliest employees.
In 2015, Baidu's autonomous driving division was established, and Peng Jun served as the chief architect of autonomous driving. Later, he met his future business partner, genius programmer Lou Tiancheng, who was also Baidu's youngest T10 employee.
At the end of the summer of 2017, by chance, Cai Chaolin, then secretary of the Nansha District Party Committee of Guangzhou, met Peng Jun during an inspection in Silicon Valley. Cai Chaolin told Peng Jun that although Nansha has not yet introduced an autonomous driving policy, it has the possibility of being a pilot area and a pilot area, and it may also be fully liberalized in the future.
This conversation greatly contributed to Peng's idea of returning to China, which included both the opportunities of autonomous driving in China, the promise of Nansha in Guangdong, and his personal desire to return to China.
In 2018, Peng Jun and Lou Tiancheng left Baidu and co-founded Pony.ai, an autonomous driving algorithm team. It was also in this year that Tesla went into mass production, and Google's self-driving project Waymo also began to hit the road. As Peng Jun expected, the global autonomous driving industry began to show the first glimmer of fire.
Almost at the same time, two other groups of people from Baidu's Institute of Fine Arts were also ready to move.
In 2017, Zhou Guang, who joined Baidu Beauty Research Institute for only eight months, also chose to leave.
Compared with Peng Jun's thoughtfulness, Zhou Guang, who had just graduated, was enthusiastic. "There is an outlet, and I'm young, so there's nothing to lose." Zhou Guang said to Interface News. After that, he founded Roadstar.ai and Yuan Rong Qixing.
In addition to old Internet people and enthusiastic young people, Wang Jin, senior vice president of Baidu, also founded JingChi Technology in 2017, attracting the former chief scientist of the autonomous driving division Han Xu to join, this company is later WeRide. Two years later, WeRide also set off to return to China and was stationed in another district of Guangzhou, Huangpu.
Han Xu has always retained the habits he had when he was a tutor at the University of Missouri in the United States. For example, after each interview, he would put on a big backpack and walk out of the venue. He is more like a university professor who has just finished a class than an entrepreneur.
In just two years or so, these scientists and entrepreneurs in the field of autonomous driving in Silicon Valley have joined forces after parting ways in the United States: to return to China to start a business and change the way humans drive that has lasted for 150 years.
The first question that these returning entrepreneurs have to consider is even the same: what path and method should be used to achieve autonomous driving? Is it to start with assisted driving and go up step by step, or to directly let unmanned driving replace the driver?
On this issue, there is a dispute between two different technical lines - the progressive and the revolutionary. Up to now, both the academic and business circles are still arguing, and passionate entrepreneurs have not been able to reach a consensus.
According to the definition of the International Society of Automotive Engineers (SAE), autonomous driving is divided into five levels: L1 to L5, with L3 as the cut-off point, the following is the assisted driving system, and the above is the high-level autonomous driving.
This distinction originated from the debate between Tesla and Google's Waymo: the former advocates gradual and gradual iteration from L2; The latter plans to start directly at Level 4 and be cut into by self-driving taxis to replace human drivers.
Pony.ai is betting on the L4 route.
"My logic is very simple, L4 is very difficult to do, but it is actually an industry with very high barriers to competition, and once we do it, the moat is very deep. On the other hand, L2 as a whole is a to B business, and it will definitely reach the stage of vicious competition, and all L2 companies will not make money today. To this day, Peng Jun has not changed his original view.
But over time, others, including Tesla, have proven that the gradual route is just as viable.
Yuanrong Qixing is a staunch supporter of gradually, turning to L2+ in 2022, cooperating with car companies to promote it, emphasizing that "end-to-end" is superior to traditional intelligent driving solutions.
Zhou Guang has the "arrogance" of being born in a robotics class. His expertise gave him the "intuition" to choose a technical route, "I knew I had to go this way," he believed, "L4 was just wrong, obviously wrong." He has publicly accused "L4 of being a lie." In 2020, Yuanrong Qixing decided to develop an intelligent driving solution that does not rely on high-precision maps, and the trial was successful at the end of the year.
Funding has become another equally critical issue compared to these technical debates.
It is much more difficult to reach autonomous driving directly than the industry originally expected, and from the perspective of commercialization, L2 can get timely market feedback and continue to generate revenue. "In 2016 and 2017, if you didn't say that you were doing L2 at that time, no one would invest." Peng Jun said that the pressure on profitability is the sword of Damocles hanging over the heads of L4 companies.
As a result, L4 players have diverged in recent years. In May 2022, WeRide cooperated with Bosch, a Tier 1 supplier, to clarify for the first time the application of autonomous driving solutions to the front-end mass production of L2 to L3.
Pony.ai chose the L2+ route. Peng Jun explained, "Because the function of the L2 sensor chip has become stronger. In the car, there can be functions similar to but not up to L4, and our advantage barriers can play a role. ”
However, Peng Jun stressed that "there has been no compromise" and that L2+ is not enough to be a moat for a company. "We are just a small product line to L2+, and the main battlefield is still L4."
But no matter how you get to Rome, the end is indeed getting closer.
In 2023, Pony.ai will start an unmanned autonomous driving test of "no safety officer in the main driver's seat" in the core urban area of Shenzhen; In January 2024, it obtained the pilot license for unmanned commercialization of intelligent networked vehicles issued by Bao'an District, and opened the country's first driverless taxi charging operation service in the central urban area of a first-tier city. So far, Pony.ai has started fully unmanned commercial operations in Beijing, Guangzhou, and Shenzhen.
Peng Jun believes that this does not mean that L4 has been successful, but at least it is technically proven that it is feasible and can be done safely, "The so-called L4 is constantly evolving, and if the proportion of unmanned vehicles in the vehicles operated by the online car-hailing platform can reach 30%, it will be very high, and it may take 5 to 10 years." Personal vehicles take longer. ”
The different technical routes represented by Peng Jun and Zhou Guang have objectively driven the great progress of the sensor industry in this land. Whether it's L2 or L4, building an autonomous car is inseparable from this component that senses the surrounding environment. The car uses it to obtain data about the external environment, which in turn allows the chip and algorithm team to make decisions about autonomous driving.
Sensors are also divided into two camps: one side supports lidar, such as Huawei Wenjie and Xpeng Motors; The other party insists on using a visual perception scheme. Tesla is a fan of the latter.
Although lidar is more expensive, it has more data. In contrast, visual perception solutions are less expensive, and algorithms can be used to compensate for the lack of camera data.
In Shenzhen, a rivalry took place between the two camps. Suteng Juchuang is the representative of the LiDAR route, while DJI firmly supports the pure visual route. The technical team was trained by Zhu Xiaorui, a professor at Harbin Institute of Technology, who is well-known in the field of robotics, among which Qiu Chunxin and Liu Letian founded Suteng Juchuang, while several other members went to DJI.
The process of obtaining investment was quite dramatic, and at first the investors could not explain what the product was when they reported to the investment committee, and finally they could only use the video of scanning the hole in the movie "Prometheus" as a reference. After the video was broadcast, everyone seemed to understand it, and finally gave hundreds of thousands of yuan in investment.
In comparison, the rise of DJI cars is smoother. In 2016, DJI set up an in-vehicle project team, but it was kept secret until 2019, when it officially decided to conduct business as a Tier1 (automotive Tier 1 supplier), and the first production car was launched in 2022, and the mass production of the Chengxing platform was launched this year.
As a direct result of this competition, the price of autonomous driving sensors has been reduced worldwide to the cost of large-scale procurement, making it possible to put autonomous driving on the road in large quantities.
In the early days, lidar was mainly produced by American manufacturers, and the unit price was as high as tens of thousands of dollars. Since 2017, Suteng Juchuang has continued to develop applications, from Robotaxi, logistics vehicles to forklifts, giving 16 lines, 32 lines, 64 lines, and 128 lines to deal with different scenarios, and realizing vehicle-grade mass production through chipization.
In the past two years, the price reduction has even given LiDAR the confidence to further challenge the visual solution. In 2021, Suteng Juchuang will bring the price of lidar below $1,000; In April this year, Suteng Juchuang released a new generation of medium and long-range LiDAR MX, which will be priced at less than $200.
The cost-effective ratio to meet the needs of car companies for intelligent driving has helped LiDAR to spread rapidly. In 2023, the total sales volume of Suteng Juchuang LiDAR products will reach 259,600 units, an increase of 355% year-on-year; In the first quarter of 2024, sales reached 120,000, a year-on-year increase of 457.4%.
The visual scheme also seems to feel the crisis. In March this year, DJI Automotive released a 7,000 yuan city pilot plan, including seven cameras and Qualcomm computing chips, without the need for lidar, millimeter-wave radar and other hardware.
Aggressive price wars are taking place. DJI Automotive expects that this solution can be installed on 150,000 yuan level models, and the lowest price will drop to 80,000 yuan. Qiu Chunchao also believes that with the mass production of MX next year, more than 200,000 cars can be equipped with lidar as standard, and about 150,000 cars can also be selected.
Although the price and cost thresholds are being sharply reduced, the debate between lidar and vision is still not over, and the market is still testing whether the two will integrate and symbiosis or go their separate ways.
The traditional manufacturing industry from Huizhou is also in the wave of intelligent driving, and they have unexpectedly become the guests of many big-name car companies.
Without intelligent driving, almost no one would have noticed Desay SV and Huayang Group. Founded in 1986, Desay SV was the first to produce traditional parts and car audio, and Huayang Group, founded in 1993, is mainly engaged in the production of video disc players.
The turning point was after 2018.
In 2019, Desay SV released the third-generation intelligent cockpit, which is equipped with a Qualcomm Snapdragon 820A "four-screen interactive" cockpit, which was subsequently procured by Ideal ONE on a trial basis.
Huayang Group has made a breakthrough in the field of rearview mirrors. Since 2016, Huayang Group has started a digital rearview mirror product line, and in the same year, it launched its first generation of streaming media rearview mirrors in China, which is the predecessor of electronic rearview mirrors. After years of product iteration and technology upgrades, at the beginning of 2022, a more mature electronic rearview mirror product will be launched.
Since then, as intelligent cockpit suppliers, these two customers have expanded rapidly in recent years, including GAC, General Motors, Volkswagen, Great Wall, and Wei Xiaoli, including almost all domestic car brands.
The reason why Huizhou can dominate the intelligent cockpit industry is that in addition to the hardware manufacturing foundation, but also because the intelligent cockpit industry chain is long and the links are complex, it can only be found in the Pearl River Delta "one-stop" enterprises that are deeply cultivated here, and a large number of companies represented by Huawei, Valeo, O-film, and Tianma Microelectronics have formed industrial synergy.
Similar to on-board radar, in the mechanical era of the cockpit, the core link of the industrial chain is almost controlled by foreign manufacturers. This can also be reflected from the background of Desay SV - Desay SV was formerly known as China Europe Electronics Industry Co., Ltd.
But driven by the rise of autonomy and the trend of electric intelligence, Chinese companies are catching up.
For example, in the field of controllers, in the era of online controlled brake ECUs, until 2022, Bosch's share is still about 90%; In the era of autonomous driving domain controllers, Desay SV ranks second, with a market share of 12%; Bosch and Aptiv have only 7%.
In addition, there is automotive head-up display technology (HUD), which was held by foreign suppliers for 80% of its products in 2021, and the share of foreign suppliers dropped to about 50% at the beginning of 2023 after Chinese companies represented by Huayang Group achieved a breakthrough.
At the same time, the "Huizhou Electronics Department" is also exporting technology in reverse. He Zhiliang, senior vice president of Desay SV, said that their company's products are ready to be exported to Europe, North America, Japan and Southeast Asia.
Huayang Group has also established subsidiaries in Germany and entered the supplier system of overseas car companies such as Honda, Toyota, Volkswagen SCNI, Stellantis Group, and Hyundai Group. In terms of precision die-casting business, it provides supporting products for overseas customers such as Bosch, BorgWarner, Continental, and ZF, and is currently promoting the implementation of overseas production bases.
Compared with the last savage wave of Internet entrepreneurship, everything seems to be more orderly in this autonomous driving track.
Autonomous driving involves public transportation and a huge R&D and manufacturing industry chain, but in 2018, Guangdong tore through policy obstacles. Today, Guangzhou has a demonstration area for intelligent networked vehicles and intelligent transportation applications, Shenzhen is an unmanned driving demonstration area, and Zhaoqing is an autonomous driving test base.
At the level of large enterprises, BYD began to wade into autonomous driving in 2015, and after the policy became clear, it has continuously increased R&D and talent investment in intelligent driving in recent years; In 2019, Huawei also confirmed that it will become a Tier1 supplier of incremental components for ICT vehicles. Xpeng Motors, on the other hand, has chosen to develop its own autonomous driving technology since 2018.
After the policy resonated with large companies, small and medium-sized enterprises also began to be agitated. Algorithm companies such as Pony.ai and Yuanrong Qixing, sensor companies such as Suteng Juchuang and DJI, and intelligent cockpit companies such as Desay SV and Huayang Group have sprung up like mushrooms in the past five years.
Autonomous driving is the holy grail of smart cars, and the realization of this disruptive technology also means that the existing automotive business model is facing a reconstruction. But at present, there are still some problems in reality: the industry is far from being able to "support a family", and high-level autonomous driving is still in the early market stage.
From January to February 2024, the penetration rate of high-speed NOA and urban NOA is about 7.62% and 3.85%, and the gap still needs to be crossed in the mainstream market.
The controversy and doubts about technology in the market have never stopped. Taking the AEB braking system as an example, this is the technical bottleneck and difficulty of the current autonomous driving technology, although it can theoretically help the vehicle to automatically brake and avoid collision in an emergency, but there are also industry views that this function has corresponding trigger conditions, when the driver has the action of turning the steering wheel in distress, or the vehicle is driving on a large curve, or the vehicle speed exceeds the threshold defined by the system, etc., AEB may not be triggered.
In addition, how much the policy can continue to open up is still a doubt about the implementation of the industry.
In order for unmanned vehicles to run through business logic, they are inseparable from three keys: first, charging operation and realizing commercial closed loop; the second is to be completely unmanned and cancel the safety officer; The third is global operation. The major premise of these three aspects lies in whether the policy is open-ended.
"Whether intelligent driving can be widely rolled out and how large the area is depends on the policy at this stage." For example, a manager of a state-owned enterprise in Huangpu District, Guangzhou, who is responsible for the collaborative construction of vehicles and roads, said that although the V2X Internet of Vehicles has been laid in the district, the collaborative application rate of equipment is only 3‰ because the policy still does not allow no one in the car.
At present, in China, Shanghai and Beijing are moving further in terms of application. By the end of 2023, a total of 415 ICVs from 15 test entities in Guangzhou have obtained road test permits, and Shanghai has issued road test and demonstration application licenses to 774 vehicles from 32 companies during the same period. As of May 2024, Shenzhen has issued a total of 1,037 road test and demonstration application notices to 349 ICVs from 19 companies. At the end of February this year, Beijing issued test licenses for a total of 384 vehicles from 18 companies.
But in any case, in the Chinese auto market, traditional foreign car companies such as Toyota and Volkswagen have dominated for decades. But right now, Chinese entrepreneurs are using technology to change the way driving has been done for a century. The wind rises at the end of Qingping, and a big wave of new technologies and industries is coming.