Autonomous driving is facing the bottleneck of "computing power", and commercialization may need to wait for another three years

Economic Observation Network reporter Gao Feichang autonomous driving, smart travel has been recognized as one of the future directions of the automobile travel industry. With the increasing popularity of intelligent connected cars and the rapid development of autonomous driving technology, many industry chain participants such as vehicle companies, software technology companies, chip companies and so on generally believe that "software-defined cars" is no longer an empty concept, but a fact.

At the recent "Chief Intellectual Officer Conference" held by Auto Byte, a platform focused on the smart travel industry created by the well-known AI information platform Machine Heart, executives from a number of industrial chain companies once again expressed the meaning of software-defined cars and put forward the saying of "software-defined cars, hardware-defined software ceilings". The main point of this is to clarify the relationship between technology and hardware coexistence rather than substitution, and to point out that the "computing power" determined by the two is a particularly critical factor in the leap from low-level automatic driving to high-level automatic driving at this stage.

Just five or six years ago, the entire auto industry was not yet mature enough to be called an autonomous car. Some models are only equipped with some functions for driving assistance, such as automatic cruise, lane keeping, lane departure warning and other functions, and the overall intelligent car is at the level of L1 and L2 in the classification of automatic driving levels. A few years later, the so-called "self-driving car" products launched by the vehicle company have been overwhelming, L2 is already standard, and L3 and L4 cars have also appeared. At the same time, a large number of new car-making forces, autonomous driving technology companies, and chip companies have emerged, and they have become the main force to promote the development of automatic driving. The guest representatives who participated in the "Chief Intellectual Officer Conference" were mainly from major leading enterprises in the industrial chain.

As one of the representatives of the vehicle company, Xia Yiping, CEO of Jidu Automobile, said at the conference that the computing power of the previous vehicle-level chip has long been lower than that of the consumer-level chip, resulting in AI technology being unable to play an advantage in the car, but the intelligent car 3.0 era can give the car enough computing power to gradually change it from a means of transportation to an INTELLIGENT mobile space driven by AI, which in turn brings technological innovation, efficiency improvement and experience subversion. He judged that 2023 will be the first year of intelligent competition in automobiles, and the real era of automobile 3.0 has arrived.

As one of the representatives of autonomous driving technology companies, Gu Weihao, co-founder and CEO of Zhixing, said that in the field of automatic driving, test mileage and test scenarios are important factors that determine the ability and safety of automatic driving systems. Data intelligence is the most fundamental driving force for the evolution of autonomous driving AI, and through further learning and mining of feedback data, the more algorithms and service models obtained by training OTA to the car side can bring better system performance to users. In this process, cost and speed are the two most critical aspects, and they are also the ideological seal of data intelligence.

As one of the representatives from the automatic driving chip company, Wang Ping, CEO of Cambrian Xingge, said bluntly that the large-scale landing of intelligent driving on the chip faces multiple challenges: the single-chip computing power is not enough, so it needs two or even multiple pieces to achieve, but this leads to a significant increase in system complexity and power consumption, increasing the cost of the system, making it difficult to popularize it on fuel vehicles or economic electric vehicles under 100,000 yuan.

He judged that there are two future trends in autonomous driving chips, one is universal openness, and the other is large computing power. In the era of L1 and L2 level autonomous driving, because the amount of data is relatively small, many car companies can accept the closed integration scheme of strong coupling of chips and algorithms, but the amount of data in the L3 and L4 eras has surged, and the algorithm is more complex, and large computing power chips are needed to meet the demand.

Based on the views of representatives of various enterprises, large computing power has become a major test restricting the development of current automatic driving, and it is also a prerequisite for the large-scale landing and further commercialization of automatic driving systems.

The test of large computing power

Hash rate is often used to refer to the performance of the chip, which is simply understood to mean that the greater the hash rate, the better the performance. With the lidar on the car, the automatic driving computing platform exceeded 1000 TOPS, and the computing power has become one of the main car selling points of more and more automobile manufacturers. However, high computing power means that in order to achieve synchronous breakthroughs in hardware and software in technology, it is necessary to retain a certain degree of redundancy, and it is also necessary to achieve a balance between technology and business in business.

In the view of Yang Yuxin, chief marketing officer of Black Sesame Intelligence, the development of automatic driving has come to the "second half of the first half", and computing power has become an important indicator to judge the degree of intelligence of automobiles, and car companies hope to highlight the computing power value so that end users have more cognition of the automatic driving capabilities of car companies. The current computing power can theoretically meet the needs of L2+ and L3 autonomous driving systems, and the next focus is to make the scene and experience better.

He also added that "computing power stacking" is a necessary redundancy for subsequent technology upgrades, from the perspective of business logic and technology evolution, chip companies also need to help customers with smaller costs, higher system concentration, lower power consumption, to achieve better automatic driving functions, which is what chip companies have been working hard to promote, but also to promote the evolution of everyone's technology evolution and product routes.

Li Bo, vice president of Lotus Technology and head of the intelligent driving business line, believes that hardware defines the software ceiling, reserves enough computing power, reserves enough sensors, and leaves redundancy for the performance requirements of future autonomous driving systems. Otherwise, it's like the current application logic can run through the old phone, but it can't really run.

In China, chip research and development is not only chip manufacturers in the current situation, in the current situation of chip shortage, the main engine factory self-developed automatic driving chip has also become a major trend, Tesla, Xiaopeng, Geely and other car companies are listed here. Wang Kai, director and CEO of Geely's Core Engine Technology, said that on the one hand, this is because the shortage of chips has made the main engine factory pay more attention to the diversity of the supply chain and supply security, on the other hand, the high-power chip has become the core competitiveness of the car company, and the supplier chip is becoming more and more difficult to meet the iteration speed, cost and performance requirements of the main engine factory.

However, Wang Kai also said that there are many challenges in the self-developed chips of car companies: the threshold of autonomous driving chips is high, and once it takes a detour, it will face huge financial losses, which will also cause uncoordinated planning. At the same time, the car grade chip is different from the consumer grade chip, the performance, power consumption and reliability requirements are higher, but also to complete the car grade certification, the cycle is longer, the investment is also larger, the need to recover the upfront cost through the application of a variety of cars, so it is necessary to launch a more inclusive and competitive product system to meet the needs of different car manufacturers.

The current chip shortage has become a huge pain point in the automotive industry, and in the face of this problem, many corporate guests expressed their views. Overall, the chip shortage will continue for some time, and although production capacity has recovered from the epidemic, the demand that was suppressed last year has not yet been met, and the real solution may not be until next year. In addition, chip manufacturers are in a wait-and-see state, the current chip expansion cost is higher, chip manufacturers do not dare to guarantee that there will be the same demand in the next few years, blindly expand production capacity.

Commercialization in 2025?

Whether it is intelligent driving or automatic driving, what it wants to achieve is automatic intelligent travel in a variety of scenario modes, so that people, cars, and transportation systems can build a fully autonomous driving society through the bond of AI technology, and change the form of human society with technology.

In the industry, the current commercialization of autonomous driving is divided into two views, one is that this is a distant dream that may never be realized, and the other is that the commercialization of autonomous driving is imminent and may be achieved in 2025. Delegates at the "Chief Intellectual Officer Conference" believe that it will take some time to commercialize autonomous driving, but some people are optimistic that projects will be implemented in 2025.

Among the popular applications of autonomous driving, Robotaxi (self-driving taxi) is the closest track to commercialization. Xiao Jianxiong, founder and CEO of AutoX, a representative company in the field, said that AutoX has been focusing on removing the L4-level driverless RoboTaxi that removes the safety officer, which believes that this is the only way to truly commercialize the route: only by achieving the same practicality of existing online ride-hailing vehicles, completely removing the safety officer, unlimited destination, unlimited area of autonomous driving, is the real commercialization.

Among them, the area covered is the most important point for Xiao Jianxiong. Xiao Jianxiong said that RoboTaxi's commercial operation must have a large enough service area, if it can only run on a few main roads, it is more of a pure technology display, and there is no real commercial value. In addition, large-scale mass production is necessary for the commercialization of Robotaxi, which determines efficiency, consistency and reliability.

The commercialization project of autonomous driving is mainly divided into two categories: B-end and C-side. Zhou Xin, co-founder and chief product officer of Yishi Technology, and Hao Jianan, co-founder and chief architect of Tucson Future, believe that efficiency and cost are the premises for the commercialization of autonomous driving at the B-end: either to achieve higher efficiency than people, or to achieve fully unmanned autonomous driving. But to achieve the final business logic, not only a very high level of safety and reliability is required, but also the gradual improvement of regulations.

As a company facing both B-end and C-end users, Dong Jian, co-founder and software algorithm VP of Hongjing Intelligent Driving, said that the current landing speed is faster than expected, and more mass-produced models will appear in one or two years. However, due to legal and regulatory issues, most car companies will launch models with L3 level automatic driving experience, but developed according to the L2+ level regulatory system.

The issue of laws and regulations on autonomous driving was mentioned by many guests, which is directly related to whether commercial projects can be approved and approved, and also involves the issue of automatic driving responsibility system.

Dong Jian said that in the past one or two years, the mass production models pushed by car companies are called L3 and L4 experience, but they are still the responsibility system of L2, which is because there is no specific L3 regulation in China, so even if the function has been L3 and L4 experience, if there is an accident in the responsibility system, the driver is still responsible. She believes that the real sense of automatic driving in L3 and L4 refers to the responsibility for the accident in the car, not the driver. Europe now has the regulation of ALKS autonomous driving in the true sense of L3, so the introduction of domestic regulations is just around the corner.

In terms of automatic driving promotion methods, many car company representatives agree that it is a relatively safe way to gradually achieve automatic driving coverage from small scenes to large scenes, and it is not in line with actual conditions to achieve L4 or above automatic driving in one step. Closed scenarios such as parks, ports, and mines are the main scenarios of current driverless applications, followed by scenarios such as trunk logistics and urban public transportation, and finally the automatic driving of personal mobility.

Dai Zhen, vice president of Heduo Technology, gave a more specific time point for the C-end landing of automatic driving - it is expected that 2025 will be a key time node, when the mass production of automatic driving technology, consumer acceptance, infrastructure and laws and regulations will gradually land.