In mid-2021, WeRide’s Mini Robobus and Robotaxi will make 500 trips to deliver anti-epidemic supplies to the lockdown area of a city in China, which also makes the public really feel to the allure of autonomous driving.
No one would have imagined that after going through a frenzied barbaric growth period and a cold capital winter, the epidemic will reshape the value of autonomous driving and accelerate the commercialization of autonomous driving.
Last year, many self-driving vehicles moved from closed road testing sites to real open roads. Robotaxi of self-driving technology companies such as Baidu, Pony.ai, and WeRide began to provide travel services to the public, and self-driving began to enter the daily life of residents. In the closed park, unmanned logistics vehicles, unmanned delivery vehicles and even unmanned sales vehicles have already started to work; on ordinary roads, unmanned sweepers have also begun to be gradually launched.
At the same time, Apple, Xiaomi, Huawei, Didi, etc. have successively announced their entry into the autonomous driving track. Wei (Lai) Xiao (Peng) Ideal (Xiang) has launched more and more assisted driving models. Traditional car companies such as Geely and SAIC etc. have also established electric vehicle brands successively, adding more self-driving technology.
But behind this prosperity, the commercialization problem of autonomous driving is also testing every player.
Autonomous driving, as the name implies, is to let the machine assist or replace the driving of the human. In this process, the degree of completion of the dynamic driving task by the machine is the level of automatic driving. According to the standards of the National Highway Safety Administration (NHTSA) and the Society of Automotive Engineers (SAE), the autonomous driving degree of intelligent networked vehicles is divided into six levels from low to high, L0 to L5. Taking L3 as the dividing line, autonomous driving above L3 (including L3) is called automatic driving (ADS), and below L3 is called assisted driving (ADAS), which is still dominated by human driving operations. The "Automotive Driving Automation Classification" issued by the Ministry of Industry and Information Technology of my country basically adopts this standard.
In 2013, Baidu first started the research and development of autonomous driving. Subsequently, autonomous driving technology companies such as Pony.ai, WeRide, and AutoX were established. Like Waymo, they focused on L4-level autonomous driving and aimed at the huge Robotaxi market. The test is also the segment that best demonstrates their L4 technology.
According to estimates by CITIC Securities, Robotaxi's potential market space is about 3.2 trillion yuan, and labor costs account for 60% of the operating costs of traditional taxis, which is the focus of Robotaxi's commercialization and upgrading. Therefore, the business path of these self-driving technology companies is to replace people with self-driving technology, launch Robotaxi products, and achieve revenue and profitability by providing Robotaxi services.
However, L4-level autonomous driving can only operate in limited areas. Generally speaking, it needs to operate in areas covered by high-precision maps and fully tested, and the opening of operating areas requires approval and support from government departments, which is also directly related to The test mileage of these self-driving technology companies, in turn, affects the commercialization process.
At present, there are hundreds of Robotaxi running in densely populated areas of Beijing, Shanghai, Guangzhou and Shenzhen to provide commercial services, but it is worth noting that the current Robotaxi is equipped with safety supervisors or remote supervisors, which is far from large-scale mass production. And fully driverless, there is still a long way to go.
At the same time, with the increase in the sales of self-driving cars, the intelligentization of cars has become a deterministic trend, attracting more and more players to enter the autonomous driving market, not only the traditional car companies GAC, Audi, Mercedes-Benz, etc., but also Weiwei (Lai) Xiao (Peng) Li (Xiang) and other new car-making forces, as well as technology companies such as Huawei, Xiaomi, and Apple, are different from the leap-forward strong technology paths chosen by the above-mentioned autonomous driving technology companies. The mass production route starts with L2 and L3 assisted driving for passenger cars, and gradually transitions to L4 and L5 autonomous driving.
According to incomplete statistics, as of November 2021, independent auto brands have released 58 L2+ passenger vehicles in 2021, an increase of more than 100% year-on-year. With the decline in the price of key components and lidar, the price of L2+ passenger vehicles has increased. At the same time, the passenger car with L3 hardware configuration (such as Xiaopeng P5) has also dropped to about 200,000, and the price is close to the people, which has paved the way for the popularization of L3 models. Yiou think tank predicts that the penetration rate of L3 autonomous vehicles will reach 20% in 2030.
With the affordable prices of L2 and L3 self-driving models, these car companies have allowed technology iteration and commercialization to proceed simultaneously, fully demonstrating the potential of the progressive mass production route. On the contrary, the large-scale commercialization of self-driving technology companies is still in the future. At present, they are still busy overcoming technical problems, accumulating test miles, and at the same time facing risks such as brain drain.
In fact, when it comes to autonomous driving, no matter what kind of technical path it is, it will eventually return to the issues of cost, mass production and commercialization.
Autonomous driving is different from traditional driving. It requires machines to complete perception, decision-making and execution like humans. To this end, advanced sensors need to be installed on the car, so that the machine can autonomously obtain and analyze driving information inside and outside the car, and continuously process some parts. Or all dynamic driving tasks.
As the first step of autonomous driving, the accuracy and efficiency of perception determine the accuracy of autonomous driving decision-making, and the sensor solution is the most important part of the perception layer. The money-burning ability of autonomous driving is obvious to all. Currently, how to use the perception solution is not only It is a technical and security issue, and there are more cost considerations behind it.
Focusing on the perception scheme, different players have two main choices. One is that Tesla is unique and chooses a camera-based visual perception scheme. "wave radar + ultrasonic radar" multi-sensor fusion scheme.
Referring to the difference between these two options, Han Xu, CEO of WeRide Technology Co., Ltd., a unicorn of L4-level autonomous driving technology enterprise, bluntly said in an interview with a reporter from Nanfeng Window that "multi-sensor fusion solutions are safer and more cost-effective". Of course, the specific The choice of the solution also has a lot to do with the level of autonomous driving, but he firmly believes that multi-sensor fusion is the best solution to achieve L4-level autonomous driving. In fact, after several years of testing, this solution has now become an industry consensus. .
It is worth mentioning that this fusion solution has been adopted by more and more traditional car companies and new car manufacturers, especially when they launched L3-level autonomous driving models, an introspection around sensors came as promised. The number of sensors in L3-level assisted driving models has generally reached about 30.
However, this also shows that the progressive production vehicle path is moving forward step by step along the preset route, and is approaching the L4 driverless level on the hardware device.
At the same time, the continuous mass production of L2 to L3 models allows the market to see the commercial value and potential of advanced assisted driving. According to my country's "Intelligent Connected Vehicle Technology Roadmap 2.0", by 2025, the L2-L3 level The sales of intelligent and connected vehicles accounted for more than 50% of the total car sales in that year, and by 2030, this proportion will exceed 70%. At present, the penetration rate of L2 is only 15%, which means that there is still more room for growth in the future.
In addition to the commercial space of advanced assisted driving, what is easy for the market to ignore is that during the operation of these mass-produced vehicles, the energy continuously feeds back data of different scenarios to car companies, which greatly reduces the data collection cost of car companies and allows them to At the same time of commercialization, it can also complete the iteration of algorithms and technologies to challenge higher-level autonomous driving, which may be the most feared thing for unmanned technology companies.
Today, Xiaopeng's plan to start the Robotaxi business, and Volkswagen's establishment of a self-driving car travel platform, all show the ambitions of traditional car companies and new car manufacturers to enter the Robotaxi track.
In Han Xu's view, the biggest difficulty in the process of scale-up and commercialization of autonomous driving is the instability of technology, and the stability of technology depends on the testing of large-scale fleets on open roads and the continuous accumulation of challenging data to improve the algorithm.
Therefore, it can be seen that in the face of the current commercialization problems, the answers given by unmanned technology companies seem to be similar: walking in a group and reducing dimensionality. Baotuan means that these unmanned technology companies themselves do not have the ability to build cars, so they choose to cooperate with car companies to form driverless fleets. For example, the cooperation between WeRide Zhixing and GAC, and the cooperation between Waymo, Baidu Apollo and Geely can be understood as preparations for scale in the future.
There are two directions for dimensionality reduction. One is the dimensionality reduction of technology. Considering that Robotaxi is difficult to commercialize on a large scale in a short period of time, these technology companies will be backward compatible with L4-level autonomous driving technology and apply it to L2+ functions. In this way, L4-level autonomous driving technology can break through regional restrictions. More open roads have been tested, and more scene data have been obtained. For example, Baidu applied its L4-level driverless technology Apollo Lite to the WM Motor W6 to achieve mass production. The second is the dimensionality reduction of the scene. The L4-level autonomous driving technology solution originally used in complex and open travel scenarios in cities is reduced in dimensionality and applied to closed scenarios with relatively simple scenarios, such as parks and mining areas. Such as the emergence of unmanned distribution vehicles, unmanned sweepers, and unmanned shuttle vehicles in the park.
However, for this kind of "dimension reduction", some people in the industry questioned that the leap-forward strong technology faction was taking measures to deal with capital's anxiety and concerns about commercialization. Han Xu, CEO of WeRide, denied this statement. He believes that the reason why the current dimension can be reduced is precisely because "now with the improvement of computing power, using the original ability to do some simple applications, of course, can do better. already."
Sixty percent of the operating cost of traditional taxis is labor costs, which means that if Robotaxi wants to make a profit, it must remove the safety officers it currently has and realize fully autonomous driving, but it is not easy to do this.
Regarding this topic, WeRide CEO Han Xu believes that in order to achieve fully unmanned autonomous driving, the biggest problem is how to achieve the stability and reliability of the autonomous driving system. Generally speaking, the first need for hardware redundancy, such as braking Redundancy of equipment, steering and other equipment to ensure that after one failure, the other can play a role, "the industry has not generally achieved full redundancy of equipment", and secondly, it is necessary to improve the algorithm in various highly complex and high-intensity scenarios stability.
Of course, he also admitted that the world does not have a completely stable system, and what they can do is to improve its reliability as much as possible, and this reliability needs to be continuously tested and verified by Robotaxi on the open road, so "I hope the government can open up the system." The more roads give us."
In the end, under what circumstances does Robotaxi need to be used in mass production? Generally speaking, what conditions must be met so that the safety of Robotaxi driving will exceed that of human beings, and the safety officer can be removed? It is recognized by the industry as the standard proposed by the RAND Think Tank in 2016, that is, under the existing algorithm framework, L4 level and above autonomous vehicles need to complete 11 billion miles (about 18 billion kilometers) of real test miles, while 18 billion The test mileage of kilometers is equivalent to 440,000 times around the earth.
But the reality is that even the recognized leader in autonomous driving, as of the beginning of 2020, Waymo's test mileage has only reached 32.2 million kilometers, and WeRide, which started the commercial operation of Robotaxi at the earliest (2019) in China, as of January 2022 On March 17, their autonomous driving mileage on public roads only exceeded 10 million kilometers, of which the fully driverless mileage exceeded 2.5 million kilometers, which is still a long way from the tens of billions of test mileage.
Insufficient test mileage means that when Robotaxi is open for operation, the probability of encountering a corner case will be higher, and the current L4-level autonomous driving is basically realized by bicycle intelligence. The risk is that even if the system can already handle 99.9% However, if the remaining 0.1% of the problems occur, it will still cause accidents, which is also the reason why the autonomous driving industry “talks about corner cases”.
Therefore, even Waymo, which has burned tens of billions of dollars in the past 10 years, is still difficult to implement Robotaxi in cities with complex road conditions, let alone give a timetable for large-scale commercial use.
It can be seen that to achieve fully unmanned autonomous driving requires both time and patience. Robin Li, chairman and CEO of Baidu, is pessimistic about this. He once said in a speech that "L5 (full unmanned autonomous driving) is too difficult and may not be realized for decades."
