Can China’s smart driving cars beat Tesla?

When Tesla FSD V13 will be released in October 2024, the topic of "who is better, us or Tesla's intelligent driving" in China is still polarized.

One reason is that since FSD has not yet entered the domestic market, many people have not had the opportunity to fully experience the latest FSD version, or their cognition has been influenced by the "open-book test" of the previous high-precision map version of intelligent driving, and they feel that "this is nothing", so they think that "our intelligent driving is far ahead."

The first group is those who are frightened by the fact that “Tesla was the first to use BEV, Tesla also has a monopoly on OCC, and Tesla is still leading from end to end.” In addition, “Tesla has the most intelligent driving cars accumulating data,” and they feel that “it will be impossible to surpass Tesla in this lifetime.”

However, can we analyze it from the underlying principles of technology? Can China's intelligent driving technology surpass Tesla? This question is suitable for the "dark horse" that is making the fastest progress and catching up from behind.

So we chose Ideal Auto. This company did not originally have intelligent driving technology as its strong point, but it has made rapid progress in the past two years. In July this year, it launched map-free intelligent driving, and as of September, it has achieved a cumulative intelligent driving mileage of 1.9 billion kilometers, surpassing Weilai's 1.041 billion and Xpeng's 982 million, ranking first among domestic automakers.

Can it surpass Tesla? With the above questions, we interviewed Dr. Lang Xianpeng, Vice President of Intelligent Driving R&D of Ideal Auto, and Zhan Kun, a senior algorithm expert of Intelligent Driving of Ideal Auto.

“Backwardness” can be reversed

If, “In the field of intelligent driving, China lags behind the United States, how can it surpass it”, can we refer to “In the field of intelligent driving, Ideal was originally lagging behind, but now it has surpassed”?

During the interview, we didn't beat around the bush and directly asked the question, "How did Ideal Auto, which lagged behind at the beginning, narrow the gap and aim to overtake others?"

"It's normal for the underachievers to counterattack and rise to the first echelon." Lang Xianpeng explained with a smile. He did not deny that Ideal Intelligent Driving had once lagged behind. After all, the "not so bright past" is sometimes not a stain, but rather can better highlight the present glory.

He attributed the most important reason for Ideal's success to high organizational efficiency.

At first glance, this seems to be a management concept and has little to do with intelligent driving technology. However, R&D is a team work that is inevitably inseparable from organization and management, and requires overall planning, coordination and optimization.

Although many consumers think that Ideal and Huawei are natural competitors, starting with Li Xiang, Ideal Auto has made no secret of its respect for and learning attitude towards Huawei. Lang Xianpeng said frankly: "We learned the organizational structure from Huawei, such as our internal IPD process. This is the result of learning some advanced experience and then internalizing it in combination with Ideal Auto's own corporate characteristics."

Zhan Kun explained in more detail that at present, the intelligent driving team of Ideal Auto is carrying out pre-research PD, R&D RD and delivery simultaneously. "We have one generation of delivery, one generation of R&D and one generation of pre-research. This is why we can always keep up with the latest technical solutions of intelligent driving. We have a relatively good step-by-step R&D process."

Just like the "coordination thinking" that Hua Luogeng popularized to everyone back then, you can "wash the teacup" while "boiling water", and two parallel things can be carried out simultaneously, and the total time is the shortest in the end.

When Ideal Auto was developing NOA, it had already pre-researched the end-to-end architecture and made preparations in advance. "So this is why we are so fast. If you think that efficiency has been sacrificed, it is actually because we have not found a way to improve efficiency," said Zhan Kun. "We have gradually found a way to efficiently verify the model through automated testing and world models, so we can balance speed and quality."

Therefore, in a sense, we can simply summarize Ideal's catch-up in the intelligent driving track as "organizational structure determines R&D efficiency, R&D efficiency determines the speed of progress, and the speed of progress determines the status of intelligent driving."

To this end, Ideal Auto has been continuously optimizing its organizational structure to speed up the development of intelligent driving technology.

At present, Ideal Intelligent Driving's R&D architecture is divided into algorithm R&D and mass production R&D, which correspond to different groups, and the groups correspond to different end-to-end modules.

According to Lang Xianpeng, in the overall strategic planning and business strategy, Ideal Auto has a clear layout for business organization. The organization changes according to the business, and the business goals and iterations are adjusted according to the strategy. This is the ideal BLM process (Business Leadership Model), formerly known as the LSA process (Ideal Auto Strategic Analysis Method).

"What everyone may perceive externally is the iteration of products and organizations, but what is actually reflected behind it is the iteration and change of our strategy and business." In his view, the changes on the surface actually correspond to the evolution of the underlying layer.

The organizational changes in Ideal Auto's intelligent driving research and development can be traced back to earlier than 2023, when Li Xiang proposed to make intelligent driving a company strategy. At the Yanqi Lake Strategic Conference in the fall of 2023, it was clearly stated for the first time that both PD and RD are very important, and they were simultaneously implemented as company-level strategies.

"Whether the organization will change in the future depends on whether it is related to the business." The essence of this answer is that "unchange is relative, change is eternal." After all, the business situation will definitely undergo major changes after an infinitely extended time axis. The company that can adjust quickly will be the one that can adapt to the market the fastest in terms of technology research and development, product creation, sales and marketing, etc.

“We can surpass Tesla”

When talking about ancient Western civilization, “Greece is always mentioned.” When talking about smart driving today, “Tesla is always mentioned.”

Greece is no longer a representative of advanced Western civilization. Can Tesla be surpassed in the field of intelligent driving? Obviously, this is considered from the perspective of technology and engineering, not from the perspective of trust in the brand.

Taking Huawei, which currently has the strongest intelligent technology in China, as an example, we believe that its advantage over Tesla is not just the word "Huawei", but "Huawei truly achieves full-stack self-developed intelligence, from the underlying OS system, sensors, software and algorithms, to the cloud, all of which are self-developed, making them easier to connect."

At the same time, Tesla’s intelligent driving is not without shortcomings.

For example, the end-to-end large model still has the shortcoming of low efficiency at the error correction end, and it is necessary for the "teacher model" to correct errors before the "student model", rather than completely severing the connection with people. Although the pure visual route has advantages such as "no redundant process of 2D to 3D conversion" and "no hindrance of multi-channel information interference", it cannot overcome the "upper limit is human driving" and the drag on the camera caused by bad weather.

Well, not all vehicle manufacturers can build an intelligent technology R&D team of 7,000 to 9,000 people like Huawei, and develop and connect all the software, hardware, systems and cloud by themselves, so how can they surpass Tesla? Ideal Auto's ideas seem to be more meaningful for reference.

"Two systems are stronger than one."

This can be a simple equation like 1+1>2 or a profound conclusion from thinking.

This is not the first time that Lang Xianpeng and Zhan Kun have introduced “System 1 + System 2”, but not many people truly understand it.

Let’s first review the end-to-end technology. This means that the intelligent driving system is no longer constrained by manually formulated rules. Instead, artificial intelligence models and mechanical self-learning methods are used to replace the perception, planning and control modules in the intelligent driving process. From the visual "input" end to the final "output" end of the intelligent driving system that controls the vehicle's self-driving, it is completely handled by the model's own system, making the functions completely black-boxed.

End-to-end model, pursuing One Model integrated end-to-end.

However, for the sake of safety redundancy, leading smart driving car companies such as Tesla, Huawei and Xiaopeng will set up certain underlying algorithms responsible for safety redundancy, making the perception and planning control modules independent, while the interfaces are still manually defined and connected.

So, the opportunity to surpass Tesla has come!

The first is to truly make the One Model fully integrated end-to-end. Ideal has done this more thoroughly than Tesla.