Author: Zhang Wei
Original: Deep Eyes Finance (chutou0325)
Level 4 autonomous driving has ushered in a critical year.
Christin Wolmar, author of "Driverless Cars: On the Road to Nowhere," has argued that the biggest stumbling block to self-driving is not public attitudes or legal issues, but something more fundamental: technology.
As mentioned above, the delay in the production of L4 autonomous driving is nothing more than the technology is not mature enough and the cost is difficult to control, such as Baidu's Robotaxi (self-driving taxi) initially cost up to 2 million units.
The cost that cannot be reduced, and the L4 enterprises that have been unable to make profits for a long time, have made the capital begin to ebb and flow. ArgoAI, backed by Ford and Volkswagen, was valued at more than $7 billion at one point and shut down last October after burning $2.6 billion; Lbeo, the head lidar company in the field of autonomous driving, has also filed for bankruptcy; Pony.ai was exposed to layoffs and Mobileye's market value plummeted.
After the capital withdrew from the L4 enterprise, it was in the cold winter and began to "get out of the virtual and into the real", which was more grounded. The most typical change is that high-end autonomous driving companies have joined the passenger car R&D business, such as Pony.ai's reduction of its Robotruck (self-driving truck) business, and the relevant R&D team is merged into the passenger car R&D business.
So, after the autonomous driving industry returned to sense, what changes have occurred in autonomous driving?
1. L4 does L2+, is serious
Following the cooling of the heat and the withdrawal of capital, the barbaric expansion stage of automatic driving has also come to an end, for the sustainable development of the enterprise in the future, making money has become the consensus of autonomous driving companies, and L4's autonomous driving companies began to "reduce dimensionality" and involve low-level fields.
For example, in addition to the Robotaxi and Robotruck (self-driving truck) business, Pony.ai launched the passenger car intelligent assistance business Personally owned vehicles, internally called "L2++".
L2++ is in line with today's capital expectations for the commercialization of L4 autonomous driving. According to public data, sales of Chinese vehicles are expected to reach 34.2 million units in 2023. Among them, the penetration rate of new cars with L2 and above intelligent driving functions will reach 82%.
This explains why L4 companies are opting for a progressive route. On the one hand, it can promote L2 autonomous driving companies to roll up and accelerate the development of L3 and L4 technologies. On the other hand, it also lays a solid foundation for L4 mass production, autonomous driving from development and design to mass production on the car, it takes about 2 years, and the emergence and development of L2+ and L2++ not only for L4 enterprise hematopoiesis, but also to support L4 mass production.
L4 is also difficult to carry in the passenger car market at present. According to data from Zhiyan Consulting, it is expected that by 2025, the penetration rate of L2 assisted driving intelligent vehicles in the world can reach 53.99%, while the penetration rate of L3~L5 autonomous vehicles can only reach 1.36%.
Unlike L4, the development of L2 and L3 autonomous driving companies is still stable, Tesla, Xpeng, NIO and other companies, benefiting from the wave of new energy intelligent vehicles, earlier formulated the development strategy of intelligent driving, layout upstream and downstream, so as to meet the demand for self-sufficiency on the eve of winter.
In the upward direction of landing, domestic L4 enterprises are more resilient and have higher ability to resist risks. The landing scenarios of foreign L4 companies focus more on a single scenario, Tucson will focus on the field of self-driving trucks in the future, and ArgoAI will focus on self-driving taxis; Domestic companies seem to be more versatile, and companies such as Pony.ai and Baidu Apollo are developing in parallel with multiple businesses such as assisted driving and unmanned driving.
L4 to do L2+, there are also many companies have divided a part of the cake, such as Moenta's mass production assistance received orders from SAIC Zhiji's main engine factory, Baidu and Lantu have also cooperated.
So, can L4 companies really reduce the dimensionality of L2 enterprises to hit L2 and L3 enterprises?
In the view of "Deep Eyes Finance", players who take the L4 leapfrog route may not be able to reduce the dimensionality to hit progressive players who are L2 and L3.
L4 companies have shown an attitude that they can at least take a sip of soup on the matter of L2+. Ford, which suffered heavy losses in the L4 winter, also shifted its focus from the delayed L4 to the L2+ front-mounted mass production. It is understood that Ford is shifting its investment focus from the long-term goal of the L4 level to the L2+ and L3 levels to obtain more direct short-term benefits.
At the end of last year, Wang Liang, chairman of the technical committee of Baidu's intelligent driving business group, said that Baidu will launch an L2+ pilot assisted driving flagship product ANP3.0 in 2023.
Unlike L4 players, progressive players such as Tesla, Xpeng, NIO, and Milli have gone through the stages of fixed-point, verification and testing, and have begun to provide high-level assisted driving products for passenger cars. According to official data, in 2022, the third generation of HPilot products have been equipped with nearly 20 models such as Weipai, tank, and Ora; Xpeng Motors has indicated that it will be the first to launch fully autonomous driving in 2023.
2. L4 enters L2+ and is no longer advanced
How to adapt high-end autonomous driving technology to L2 models is the difficulty of L4 entering L2.
Yuan Rong Qixing ECO Zhou Guang once said: L4 companies do L2 two ways.
"The first is to rely on independent teams and independent code to achieve specific subdivision functions of L2, but this will make L4's technical accumulation less useful; The second is to reduce the demand for sensors and computing power for L4-level software system diagrams, and stuff the L4 software system computing power and power consumption into the L2 hardware, but the L4 automatic driving system has high computing power and power consumption, and only a few companies can put it on the vehicle-level computing power platform. ”
For the functions required by L2+, L2 car manufacturers are not far behind. Tesla, ideal NOA, Xpeng's NGP, NIO's NOP, and Momo Zhixing's NOH, although the names are different, but they are all what we call L2+ in a broad sense.
The technological advantages of L4 enterprises have not reached the level of "far beyond" the technology of L2 enterprises.
The lack of lidar and computing power was once the main technical problem of the L2 track, and with the rapid development of L2 enterprise hardware in the past two years, the situation has greatly improved. On the one hand, L2 new cars have begun to be equipped with lidar, such as NIO ET7, Xpeng P7, WM M7, Zhiji L7 and other models are equipped with lidar. On the other hand, with the emergence of high-computing power chips such as Orin and Horizon J5, computing power is no longer a bottleneck, such as Qingzhou Smart Navigation and Horizon cooperation to launch Horizon Journey 5, in October last year, ideal has been equipped with Horizon Journey 5 on the main model.
Huang Chang, co-founder and CTO of Horizon, also said that he will launch a high-end and more powerful journey 6 to compete with NVIDIA and Qualcomm. It is reported that Baidu's L2+ assisted driving product ANP3.0 will be launched this year, which is equipped with two NVIDIA Orin-X chips.
Most L4 companies have data closed-loop capabilities and high-level R&D capabilities, but lack engineering capabilities, and their advantages in sensors, domain control, integration and other aspects are not obvious enough, so L4 companies choose to cooperate with some traditional big-name manufacturers, such as WeRide and LiDAR manufacturer Bosch, to jointly promote the application of L2-L3 autonomous driving large-scale pre-installation mass production and market-oriented applications.
Although the technical advantages of L4 enterprises in L2 are not obvious, the time for L4 enterprises to develop L2 products can be greatly shortened from R&D to test landing and market entry.
Previously, the road scenarios targeted by L4 were mainly urban roads, while the traditional L2 only had a small number of functions involved in urban roads, such as traffic congestion assistance, and more for highway scenarios.
As the trend of electrification and intelligence of passenger cars becomes clearer, L2 begins to develop on a more complex road, such as urban NOA scenarios and L4 begin to approach, which means that L2 and L4 begin to converge more and more, so the R&D system, organizational structure and testing methods behind L2 and L4 companies begin to coincide.
L4 squeezes into the L2 market, the performance of the system and the understanding of applicable scenarios are different, the degree of customization required is high, and customization will increase the cost, increasing the demand for the collaborative ability of the industrial chain.
From the perspective of L2 system, assuming that L2 or L2+ systems are developed by several suppliers, the production of the host is actually completed by multi-party cooperation, whether it is logical rationality or smooth operation.
3. Mass production, L4 future gamble
"Now no investors are willing to listen to the company's talk about BP, and prefer to interview downstream scene customers alone to understand the company's real mass production profit prospects." This is an investor's feedback on autonomous driving.
Mass production is the most important step in the commercialization of autonomous driving. The emergence of L2++ also shows the importance of mass production routes in automatic driving, and Tesla, Milli Mobility and other companies have confirmed the feasibility of mass production L2 intelligent driving.
In July 2020, Li Auto launched the banner of L2 mass production and self-research, and the new forces of car manufacturing began to develop rapidly, and NIO simultaneously built a self-research team; Tesla prepared the DOJO Intelligent Computing Center for FSD; Backed by the Great Wall, the defender behind the city's NOH has been pushed to the front of the stage.
Coincidentally, while promoting the large-scale landing of Robotaxi, Baidu is also actively promoting the mass production of Apollo pilot assistance.
In mass production intelligent driving, cost is an important factor in competition between enterprises, and it is also a major problem for enterprises to carry the burden forward. Zhou Guang, CEO of Yuanrong Qixing, has said many times: In the end, autonomous driving must be a low-cost, highly automated solution.
Theoretically, the 300,000-yuan mass-produced passenger car, relying on massive data and efficient technical closed loops, is significantly superior to Robotaxi in terms of autonomous driving landing speed and scale.
In the past, L4 was always difficult in cost, and the cost of the shared unmanned car Apollo Moon released by Baidu Apollo and Beiqi Extreme Fox the year before was as high as 480,000 yuan, which was still one-third of the average cost of L4 autonomous driving models in the industry at that time, which shows the high cost of L4 autonomous driving.
At present, some companies have knocked down the cost, such as the L4 unmanned vehicle Apollo RT6 launched by Baidu last year, with a cost price of only 250,000 yuan; Qingzhou Smart Navigation released the DBQ V4, a vehicle-level autonomous driving solution with a mass production cost of only 10,000 yuan.
If the cost has been cut down, why is the company slow to make a profit? The reason why the cost of Level 4 autonomous driving has been knocked down is not the maturity of the L4 industry, but the castration of high-cost technology by enterprises.
Tesla, for example, has gone to great lengths to reduce vehicle costs. Tesla first abandoned millimeter-wave radar, and canceled ultrasonic radar last year, and as a result, Tesla Model S and Model X directly stopped using radar and switched to 100% pure visual autopilot.
Autonomous driving itself is a relatively large systematic field, including multiple fields, not only requiring technical cooperation, but also involving traffic conditions, data positioning, safety and other factors, such as high-precision maps, road coordination.
The research and development stage is the process from 0 to 1, at this time the investment is the imagination and vision of the future, and mass production is a leap from 1 to 10, which is the landing link, and the feasibility of the product in the market should be considered.
The development of L4 is limited by the overall development level of the autonomous driving industry. In 2017, L2 did not receive much attention, and even Tesla's Autopilot did not many people think it was very remarkable, until the self-driving vehicles of the new car-making forces began mass production, and the industry began to accelerate its development.
2023 is a key node for high-end assisted driving, and it is also the first year of high-end assisted driving. Ni Kai, CEO of autonomous driving company Heduo Technology, also believes that the next three years may become the window for the entire high-end autonomous driving market. The large-scale landing of high-end autonomous driving will enter a mature period in 2025, and the industry trend will become linear or stable growth.
This year's L2+ mass-produced models will be launched one after another, and L2+ as the wind vane of L4 mass production, the L2+ market has become a key competition point.