Tesla FSD will enter China this year？ Is FSD a benchmark or drawing a cake？

Text |shanghai auto news

The recent recall of Tesla was interpreted by many people as that the relevant departments may be preparing for the release of the FSD (fully automatic driving function) that Tesla is proud of in China.

At present, Tesla has not opened FSD in China. Previously, there have been many rumors that Tesla FSD will be released in China. Chu Ruisong, general manager of Baidu Smart Car Division, believes that FSD will enter the China market this year or next year, and it will be opened on a large scale in 2025. Recently, more "signs" show that FSD will enter the China market this year. A few days ago, the relevant person in charge of the Intelligent Manufacturing Promotion Department of the Shanghai Economic and Information Committee introduced that in the next stage, Shanghai will further deepen its cooperation with Tesla and promote the layout of functional sectors such as autonomous driving and robots in Shanghai.

If this catfish, which represents the intelligence of automobiles, is really introduced into the China market, can it produce the catfish effect again?

Before answering this question, it is necessary for us to discuss: Is FSD the benchmark of the autonomous driving industry, or is it simply drawing cakes?

The argument has never stopped, and the real reason lies in "value for money"

The controversy surrounding FSD begins with its name.

Tesla FSD literally means Full Self-Drive, which is actually an upgraded version of the automatic assisted driving function. FSD is not installed on the vehicle to realize fully automatic driving, but it can be realized through wireless software update later. FSD is a fully automatic driving option launched by Tesla. Compared with the Autopilot function of Tesla’s current loading application, the function of FSD is more advanced, and it can realize the functional requirements such as automatic parking, automatic lane change assistance, traffic signal light recognition and automatic response, intelligent call of parking lot, etc., and can realize the "automatic driving" function through software subscription.

With the advent of FSD, American consumers have been questioning it constantly, and relevant departments in the United States have also investigated it many times.

Perhaps it is more convincing to use the exaggeration of Apple co-founder Steve Wozniak in an interview with CNN recently: "If you want artificial intelligence to make mistakes and get a lot of claims, you will try to kill you every chance, then you can buy a Tesla."

Then, since everyone is an assisted driver, apart from the name dispute, why are other models not as controversial as Tesla?

In addition to Tesla’s large number of vehicles and intelligent driving function is its main selling point, to put it bluntly, it is still a question of money. Tesla FSD is available as an option, and the autopilot subscription was released in the United States last year. The business model of Tesla’s automobile business further extends from sales revenue. Through the software charging model, a car can contribute revenue to the company throughout its life cycle, and FSD is the company’s most important payment software.

Do you understand? If you charge a lot of money, you must accept the responsibility of consumers to be more picky.

Is FSD the "ceiling" of domestic autonomous driving?

In the automotive industry, Tesla first applied Transformer to the intelligent driving of production vehicles, one to two years earlier than other companies. Transformer is a neural network model that makes the whole AI industry boil (Transformers). Prior to this, the perceptual algorithm module of assisted driving mainly ran convolutional neural network CNN. In 2022, Tesla launched an Occupancy Network that can detect general obstacles at the new AI Day.

In short, the advantage of Tesla scheme is that it can make judgments and decisions faster without relying on high-precision maps. The core idea of evolution is to be faster and more accurate.

Tesla’s system will hardly be "broken". Many testers’ videos on the network are shot in complex urban road scenes. Some road conditions are as complicated as big cities in China, but it is obvious that Tesla FSD has made progress.

At present, many independent brands have realized the functions of automatic lane change, automatic overtaking and automatic on-off ramp, and the whole driving process is very coherent. The two are different in implementation. Tesla has always advocated the vision radar scheme, while the current mainstream practice of independent brands is to achieve high-order driving assistance through laser radar and high-precision maps. At the same time, these car companies will also cooperate with visual radar and millimeter-wave radar to make up for some scenes that lidar can’t cope with. However, because it relies heavily on high-precision maps, the high-order driving assistance function will not be realized after losing high-precision maps.

In order to get rid of the heavy dependence on maps, companies such as Ideality, Tucki and Huawei have begun to actively develop high-level driver assistance functions with heavy perception. From this point of view, these companies have more or less defaulted to Tesla’s "master" role.

According to the latest news released by BYD, the new head of BYD’s intelligent driving is in place, directly aiming at the high-end high-speed navigation function and "benchmarking" the Tesla FSD experience. The high-speed NOA, the first product after the adjustment of BYD’s intelligent driving system, is said to be mass-produced and listed this year. The most important thing is the core algorithm, which is developed by BYD, not purchased.

FSD, once again as a catfish?

Once upon a time, Tesla, as a new energy vehicle, was introduced into the relatively closed traditional automobile market in China. As a result, with the policy blessing, the deduction of various kinds of capital and the change of consumers’ concept, no one could have imagined that this catfish would make new car-making forces rise one after another, and the penetration rate of the new energy vehicle market increased by leaps and bounds, which even caused great changes in the automobile industry in China and the world.

Nowadays, autonomous driving is facing a dilemma: L4-level autonomous driving is far away, and many enterprises that originally specialized in L4-level autonomous driving began to descend to L2 or L2+.

The reporter believes that if the government wants to find an opportunity to "break the game" in the field of automobile intelligence, re-find the power point and make autonomous driving accepted by ordinary consumers faster, it is a good way to let Tesla FSD "catfish" enter the market as soon as possible.

However, the problem is that it is difficult for Tesla’s FSD to regain the glorious course of Tesla’s first entry into the China market. There are three main reasons: First, when the new energy vehicles represented by Tesla just emerged, the sound of optimism was basically higher than that of pessimism around the world. I remember that in addition to the energy crisis at that time, many people bought Tesla new energy vehicles, which represented a fashion life attitude and a public interest in environmental protection. At present, FSD has been questioned all over the world.

Second, the development of new energy vehicles can be said to be promoted by the full policies of governments in various countries, especially in China, which has great support and the most outstanding achievements. However, at present, due to regulatory difficulties in autonomous driving, governments of various countries have different attitudes, and China may gradually encounter regulatory difficulties with the rapid growth of the number of consumers participating in autonomous driving applications.

Third, in terms of technology, capital and even regulations, can Tesla’s FSD really make people abandon painting cakes?

Painting cakes? Still have enough tolerance for it.

The research and development cycle is long, it is difficult to land, and the return is far away … Autopilot companies and their shareholders have to re-examine the benefits behind them. Especially in the context of economic downturn, limited supply chain and rising raw material prices.

Doug Field, Ford’s chief senior product development and technology officer, said that it is "more difficult than landing on the moon" to achieve autonomous driving in a dense urban environment. Wang Chuanfu, the boss of BYD, said: "Now the so-called’ automatic driving’ is held hostage by capital. I think that at the end of the development of autonomous driving, at most,’ advanced assisted driving’ can be achieved, and that’s all. When you are sleepy and reading short messages, reduce the probability of accidents. There will definitely be accidents and injuries, but the responsibility must always be borne by the owner himself. "

Transformer has rigid requirements on data size. If the data size does not meet the requirements, the performance will be worse than CNN. Therefore, the premise of switching to Transformer is that enough data can be obtained to feed it.

In this regard, Tesla has greater advantages as a car company with a high number of intelligent vehicles, but this is only the beginning. Because it is one thing to collect more and higher quality data, it is another to use them to train a strong enough model. Training a large model of massive parameters in massive data requires a lot of computing power, and only a large cloud computing center can be competent.

In 2022, in order to occupy the network for training, Tesla used 1.44 billion frames of video data, with a data volume of more than 30PB, and its self-developed DOJO supercomputer is still in the development state.

Moreover, at present, almost all chip factories provide smart driving chips on the car side, which are not prepared for Transformer.

It is a fact that even the breakthroughs such as Transformer, BEV and occupied network, which make the industry chase after each other, are still mainly solving the perception problem of intelligent driving vehicles, and there are still difficult decision/regulation links in the complete intelligent driving technology framework.

From the data, computing power, chips to the legal level, we can roughly draw the conclusion that even if it is as strong as Tesla, it is obviously painting a cake if we want to realize the true FSD (full automatic driving L5 level), and what we can do at present can only be an infinite "+"after L2 level.

Of course, it may also bring about the final change, but I don’t know the specific time.

For consumers, on the premise of ensuring safety, an infinite plus sign after L2 level means a better consumer experience.

The question is, how many enterprises can wait in the face of the journey with no end in sight? Do they really need to cope with this competition?