Shi Jie, Automobile Commune: The "root system" of smart cars comes from the development capabilities of the underlying operating system

Our reporters Yin Limei and Tong Haihua report from Beijing

"Automotive intelligence can provide us with a lot of functions and cool technologies, but it is inseparable from the support of some underlying applications and architectures, such as basic software. Many developers are reluctant to develop basic software because it is far away from the terminal, but the basic software is a decisive factor in determining whether the intelligent function will work effectively. If we don't do a good job with the basic software, then the final effect of the application of intelligent functions will also be very significantly dragged down. ”

Recently, Shi Jie, editor-in-chief of the Automobile Commune, made the above statement in an exclusive interview with reporters in a live broadcast jointly organized by the China Association of Automobile Manufacturers, China Business News, and China Economic Media Think Tank.

The basic software, represented by the operating system, is the key to connecting the underlying computing power and the upper-layer applications, and is the soul of the digital infrastructure. Recently, there has been a lot of discussion about basic software in the industry.

Shi Jie told reporters that for a long time, consumers and people in the automobile industry have focused more on the "superstructure" of automobiles. Although many vehicle companies have launched their own car machine systems, the actual situation is often that they just "do some processing" on the existing underlying basic system, carry out some customized development, and finally name their own brands, these car companies are not really the underlying basic system developers.

For example, Shi Jie said that as far as the Hongmeng system developed by Huawei is concerned, the early Hongmeng is not a pure Chinese self-developed operating system, it has a certain relationship with the Android system, and can be compatible with APK applications using the Android system. However, the latter HarmonyOS is an operating system independent of the Android system and is developed based on the microkernel system. Since HarmonyOS, China has had an underlying operating system that we have developed completely independently from beginning to end.

In Shi Jie's view, the mainland can derive a consumer electronics operating system on the underlying operating system, as well as a car machine and intelligent driving operating system. Only with the ability to independently develop the underlying operating system can our industry be like a big tree with a developed root system.

In fact, in recent years, the industry has been paying more and more attention to the independent innovation and research and development of basic software, and building an independent and controllable open source operating system for intelligent networked vehicles. At present, there is a common understanding in the industry that open source will be the most efficient way to develop operating systems and basic software.

The reporter noticed that Ni Guangnan, an academician of the Chinese Academy of Engineering, recently called for attention to the open source RISC-V architecture in public.

As a brand-new, open-source instruction set architecture, RISC-V has been joined by more than 4,000 members and tens of thousands of developers in more than 70 countries around the world, and the number is still growing.

"RISC-V will bring new opportunities for the development of China's basic software." Ni Guangnan believes that although the application of RISC-V in the field of intelligent networked vehicles is still in the early stage and has not yet been put on a large scale, based on the technical characteristics and open source model of RISC-V, RISC-V chips have broad market prospects in the future, which will surely drive the high-quality development of the intelligent networked vehicle industry and become a new quality productivity in the automotive field.

"Why did Academician Ni Guangnan and many research institutions and consulting companies mention that by 2029, about a quarter of the world's chips will use RISC-V architecture? What's the logic behind this? The reason for this is that RISC-V is an open-source architecture. Relatively speaking, under the RISC-V architecture, we will not be coerced by intellectual property rights and will not face high costs; At the same time, under the RISC-V architecture, the probability of us being 'stuck' is relatively small. Shi Jie said.

In addition to the operating system and basic software, the "car" on the large model is also a topic of considerable concern in the automotive industry, and the industry is exploring the use of end-to-end large models to achieve high-level intelligent driving.

At present, the intelligent driving system is generally divided into three modules: perception, prediction, and planning, and the end-to-end model integrates the three modules, inputs information from the perception end, and directly outputs the results at the execution end, so it is called "end-to-end".

Shi Jie told reporters that although the "end-to-end" technology path was popular this year, it is not actually a term and concept that only appeared this year, and Tesla is not the "first to eat crabs" company, Toyota has tried to take the "end-to-end" technology path a long time ago.

"In intelligent driving, the traditional approach is that many modules involve rules that need to be artificially defined individually, but 'end-to-end' is to let the system learn, train and build rules by itself, and the large model is a step forward on top of deep learning." End-to-end' has advantages and disadvantages. The advantage is that we don't need to use more complex interventions, and it will be more efficient. But the disadvantage is that this system must go through a lot of self-learning and self-training, and the investment and cost will be very high. Shi Jie said.

Shi Jie said that the reason why Toyota did not vigorously promote "end-to-end" technology later was because this path required a high cost. Toyota regards intelligent technology as the "icing on the cake" function on car products, while Tesla regards intelligence as the life of its entire car brand. Due to different ideas, Toyota, as the pioneer of "end-to-end" technology, did not make full use of it, but Tesla took the lead and made this "end-to-end" technology bigger.

In fact, in addition to the higher cost, there are certain risks associated with the "end-to-end" technology path. Shi Jie told reporters: "At present, some people in the industry are questioning, no one artificially formulates rules, how can it correct deviations and errors? This issue is still being discussed. 'End-to-end' technology is constantly growing and improving, but it needs to be developed in terms of cost, time and technical difficulty. ”

It is worth noting that in recent years, the momentum of mainland vehicles going overseas has been fierce, and topics such as automotive data security and smart cars going overseas have also attracted much attention.

Shi Jie said that how to effectively integrate with the local regulatory system and user habits has always been a topic of concern about how cars go overseas. In the era of automotive intelligence, the situation involved will be more complicated, because intelligence will give the automotive industry more commercialization possibilities and opportunities to reflect value, but also bring more challenges and risks.

"Smart cars generate a lot of data, including user usage data, trajectory data, and some manufacturing data. How should the ownership and use rights of the relevant data be defined? How can we better adapt our technology to the regulations and standards of the target market? It's going to take a very long time. There will be some twists and turns and risks in this process, but although it is difficult to do, it will be done, and we should not retreat from the difficulties. Shi Jie said.

(Editor: Zhang Shuo Review: Tong Haihua Proofreader: Yan Jingning)