Autonomous driving is difficult to attack by dimensionality reduction

Since 2021, some L4 autonomous driving companies targeting Robotaxi have begun to differentiate into new businesses, trying to make breakthroughs in commercial revenue through technology.

The phenomenon of autonomous driving companies developing L2-level assisted driving functions is called a "dimensionality reduction attack" by some people.

In their view, L4 autonomous driving companies can rely on the advantages of software algorithms to provide smart cars with richer functions and driving experiences, just like the high-dimensional creatures in "The Three-Body Problem" that can easily destroy low-dimensional creatures and Civilized ease.

However, some people are dismissive of this. They believe that it may not be a smooth process for autonomous driving companies that are good at software algorithms to enter the research and development field of automotive functions that emphasize the combination of software and hardware.

Both sides hold their own opinions and it is difficult to distinguish. But what is certain is that autonomous driving companies entering the increasingly competitive automotive supply chain may contribute new technological capabilities to the development of smart cars.

Is it a smooth road for L4-level autonomous driving companies to embark on the forked road of developing L2-level assisted driving?

Choices under Crisis

Many people believe that shopping malls are like battlefields. In the endless business war, if companies can make good use of the art of war, it may help them get out of trouble.

This situation also applies to some autonomous driving companies trapped in the fog of commercialization. They are looking for a surprise attack to help them break through the blockade.

While thinking, they looked at the L2 assisted driving system.

In the past period of time, some L4 autonomous driving companies have focused on the Robotaxi business at the top of Mount Everest, taking it as their mission to realize the commercial scale operation of fully unmanned Robotaxi and climbing to the summit at any cost.

As a few years go by, the distance between them and the end point is gradually shortened, but the target is still out of sight, and the consumption of funds is increasing as the scale of autonomous vehicles expands.

A supporting example is the recent second-quarter financial report released by General Motors, which revealed that Cruise, its autonomous driving company, was losing more than US$5 million every day, and its losses increased from US$600 million in the same period to US$900 million.

Cruise Autonomous Driving Fleet

At the same time, with large-scale commercialization in sight and the economy cooling, investors’ supplies are gradually decreasing. . In the crisis of lack of food and clothing, some companies collapsed due to exhaustion on the road, while others were seriously injured due to lack of oxygen and cold.

At this time, their primary goal is no longer to look up at the sky, but to survive.

Fortunately, changes in China’s automobile market have provided self-driving companies with survival opportunities.

On the one hand, The rapid development of smart cars in China requires the integration of autonomous driving technology.

Under the intelligent assisted driving system promoted by Tesla and Xpeng Motors, whether mass-produced cars are equipped with high-level automatic driving functions is becoming one of the biggest selling points of the product. In order to avoid falling behind in technology , Great Wall, Geely, Volkswagen, General Motors and other OEMs have established autonomous driving departments to develop autonomous driving functions at L2 and above.

However, the development of autonomous driving systems is an option for a few OEMs, and currently more OEMs have temporarily given it up due to limited conditions. In order to gain more market attention for subsequent products as soon as possible, whether it is self-researched or not, it is a good choice to seek external cooperation to develop an autonomous driving system that can be mass-produced.

On the other hand, In the context of the rapid development of the smart car market, domestic OEMs have begun to tend to adopt more flexible cooperation methods, and the complete solution delivery methods of foreign traditional suppliers have gradually It does not meet the needs of OEMs, which provides opportunities for domestic suppliers to fill positions.

In addition, most of the traditional foreign suppliers have previously set up their technology research and development centers in their own countries, which makes it difficult for their technologies to meet the rapidly changing needs of the Chinese market.

Monitoring data from Gaogong Intelligent Automotive researchers show that as of the end of the first quarter of 2022, Bosch's market share in China's passenger car front-end standard equipment market has dropped from 27.79% in 2021 to 25.61%.

The relative lack of localization of traditional foreign suppliers has led OEMs to try to cooperate with domestic automotive suppliers, which has contributed to the growth of the latter. A number of local Tier 1 companies such as Moshi Intelligent and Freetech have grown rapidly.

Inwardly, it is difficult to commercialize L4 autonomous driving technology, and we need to find alternative ways; outwardly, market demand is strong and the momentum of powerful enemies is weakening. It seems natural for autonomous driving companies to develop L2-level assisted driving systems.

More importantly, as the computing power of smart cars increases and the number of sensors increases, the overall cost drops significantly, providing conditions for the low-cost mass production application of intelligent assisted driving systems.

In the midst of an existential crisis, L4 autonomous driving companies took advantage of the situation to change course and found a possible way to survive.

Is it really a dimensionality reduction attack?

Whether this road is smooth or not, no one knows, and it doesn’t matter. Importantly, some self-driving companies know that there seems to be a glimmer of hope in going down this path.

Is this really possible?

Wang Hua (pseudonym), an autonomous driving practitioner, told Xinzhijia that there is no necessary connection between the assisted driving system of the L2-L3 level system and the fully unmanned system of the L4-L5 level. The assisted driving system can be upgraded to Autonomous driving, but the autonomous driving system cannot directly reduce the dimensionality to assisted driving.

If there is no necessary connection between the two parties, what difficulties will L4 autonomous driving companies face when developing L2-L3 assisted driving systems?

Technical difficulties are an unavoidable hurdle.

First, autonomous driving companies have relatively little experience in mass production of passenger cars.

Many ADAS suppliers told Xinzhijia that most L4 autonomous driving companies do not care about the cost of technology research and development, and usually use high-power chips and do not pay much attention to the technical route of mass production. If they switch to L2-level assisted driving, it may be difficult to "transition from luxury to frugality", for example, they will be constrained in product development. More importantly, dimensionality reduction does not mean that the algorithm can be reused, and many things need to be overthrown and rebuilt.

In the past, autonomous driving companies used software algorithms based on deep learning as a moat to compete for the stability and advancement of autonomous driving systems, paying less attention to how to pre-install autonomous driving systems in mass-produced cars, resulting in Self-driving companies don’t know enough about car manufacturing.

Yin Wei, senior manager of Zhiji Software, told Xinzhijia that the cooperation between OEMs and autonomous driving companies can also be said to be learning from each other. When autonomous driving companies encounter problems when dealing with vehicle products, they will learn from OEMs how to solve them; OEMs will also learn good methodologies from autonomous driving companies and strengthen their software development capabilities.

Second, Autonomous driving companies have relatively insufficient data accumulation in some scenarios.

The technical person in charge of an autonomous driving supplier told Xinzhijia that there are differences in the application layer between autonomous driving in low-speed scenarios and high-speed scenarios.

Autonomous driving requires long-term testing to collect data and then iterate the system. The difference between low-speed scenarios and high-speed scenarios leads to different data collected by autonomous driving in two different scenarios. In the end, the practice methods of the two are also different.

In the past, self-driving companies focused on collecting different types of data in high-speed scenarios, which are widely applicable, while companies that started developing assisted driving functions focused more on low-speed scenarios.

Completely different scenarios, and of course the data are also completely different.

Therefore, L4 autonomous driving companies do not have an advantage in developing functions for low-speed scenarios such as autonomous parking.

Third, Developing an autonomous driving system when computing power is greatly reduced is like doing a dojo in a snail shell.

Many self-driving companies have often mentioned their high computing power when promoting the responsiveness of self-driving cars in the past. Thousands or even more than 2,000 Tops of computing power seems to have become the standard for self-driving systems. However, mass-produced cars Due to cost considerations, it is currently impossible to configure chips with higher computing power.

For autonomous driving companies that are accustomed to developing under conditions of large computing power, how to complete the development of L2/L3 autonomous driving systems under extremely limited computing power is a huge challenge. Some people have expressed doubts about the new Zhijia vividly compares it to "the difficulty of transitioning from luxury to frugality".

The road to dimensionality reduction for autonomous driving companies is long and difficult.

Non-technical problems of self-driving companies

If the self-driving company solves the technical problems, the process of getting the assisted driving system into the car may not be easy.

First of all, autonomous driving companies need to realize low-cost mass production of solutions.

In the past, some autonomous driving companies have focused on technology and tried to create as reliable a perception system as possible to achieve autonomous driving through a large number of sensors of various types. However, this has resulted in higher costs for autonomous driving solutions. , the autonomous driving solutions released by many companies cost hundreds of thousands or even millions of yuan and cannot be applied on a large scale.

If an autonomous driving company tries to reduce the overall cost and apply it to mass-produced models, it needs to consider how to reduce hardware costs while meeting vehicle regulations. This is a problem for autonomous driving system developers who are less likely to deal with hardware. It's really not easy.

After autonomous driving companies launch low-cost mass-produced autonomous driving solutions, whether the OEM can foot the bill is also a question.

"Any car company that does not develop autonomous driving will die." Wang Xiaoqiu, President of SAIC, said this, reflecting the importance of OEMs to independently control autonomous driving technology.

In order to control the software, foreign OEMs such as Ford and Toyota acquired autonomous driving companies earlier, while Volkswagen established an autonomous driving department.

Yin Wei told Xinzhijia that Zhiji Auto prefers self-research and open source cooperation with suppliers.

As domain controllers and OTA increasingly determine vehicle performance and driving experience, OEMs realize that in the development of smart driving functions, they must better integrate self-research and open source cooperation within one controller Combined.

This means that if the OEM cooperates with an external autonomous driving company, the autonomous driving company may not deliver a complete set of autonomous driving solutions, but conduct customized joint development for related functions.

When autonomous driving companies cooperate with OEMs, whether they can deliver products as scheduled may also be a problem.

An industry insider told Xinzhijia that if the two parties only cooperate once or twice, the autonomous driving company can provide hundreds of customized autonomous vehicles to the OEM in a short period of time, and the product functions are very It is difficult to completely satisfy the OEM.

He believes that a truly mature product requires long-term cooperation between both parties, because the product will have autonomous driving technology, wire control technology, and new software technology, which will require a long period of mapping and product polishing to complete.

At present, although some autonomous driving companies try to jointly develop products with OEMs, both parties will inevitably encounter many problems during the cooperation process, which may delay product delivery.

Searching for another commercialization path for autonomous driving companies may involve dimensionality reduction, but this process is not easy.

Changes, the new theme of autonomous driving

In recent years, as many autonomous driving companies have successfully obtained commercial revenue, there have been widespread rumors that the autonomous driving industry is about to enter the knockout phase.

A group of autonomous driving companies have taken flight, while another group is still confused.

If focusing on a certain scenario was the secret to the rapid development of autonomous driving technology in the past, then getting rid of the boundaries of self-restraint and truly practicing the technology in different forms may be a new theme in the future.

L4 autonomous driving companies are turning to the development of assisted driving functions, which is a new force entering the automotive supply chain and contributing to the development of China's automotive industry.

But it is undeniable that autonomous driving companies have just started to develop assisted driving, and there are still many difficulties ahead.

Looking forward to L4 autonomous driving companies making new progress in the mass production of assisted driving functions.

The above is the detailed content of Autonomous driving is difficult to attack by dimensionality reduction. For more information, please follow other related articles on the PHP Chinese website!