Author |Edited by Zou Yuyuan |Produced by Fan Qianglong |In the last few months of 2023, car companies have not only gone crazy in the first battle, but also reached a new height in intelligent driving: many traditional car companies have obtained L3 autonomous driving road test licenses in mid-to-late December, and new forces have started the collective landing of full-scene assisted driving. Ideal NOA, Xpeng NGP, and Huawei NCA, the most familiar intelligent driving systems, have begun to really land in the most common urban scenarios in our daily life. Even the just-released Xiaomi car has also demonstrated its intelligent driving strength.
When the issuance of L3 license plates and the "collective submission" of the new forces are superimposed, it gives us a feeling that at the end of 2023, the intelligent driving industry has taken a big step forward. It is this leap that has formed a strong difference between the entire automotive industry, and the Matthew effect is spreading from the field of electrification to the field of intelligence, with the strong becoming stronger and stronger, and the weak not even having the opportunity to be on the table. 
There is a huge difference between the old and new forcesThe new forces we are talking about here include not only Wei Xiaoli, Leap, Nezha, but also the new brands incubated by traditional car companies.
In terms of high-end intelligent driving, all companies invariably put a variety of high-end perception hardware on the basis of NVIDIA Orin chips, such as multiple 8 million pixel lenses, various forms of lidar, 4D millimeter wave radar, etc. 
After the development of 2021-2023, everyone has basically landed high-end intelligent driving in the city, and in order to adapt to more diversified scenarios and achieve the goal of reducing costs, it has almost become the consensus of the entire industry to dilute or not rely on high-precision maps to do intelligent driving. Of course, this is also accompanied by the rapid development of AI artificial intelligence, large models and other technologies, so that new forces can quickly switch architectures, so as to accelerate the implementation.
Nowadays, with the full-stack self-developed Xpeng, Ideal, and Huawei systems, the level of intelligent driving can be said to be at the head of the industry. In 2023, the vast majority of other new forces will still be at a level that can only achieve high-speed navigation assistance. In fact, this does not reflect the difference between the high and low that we are talking about, but this huge gap can be seen between the new forces and traditional independent brands and joint venture car companies.
Of course, they have certain autonomous driving research and development capabilities in overseas or local markets, such as Nissan, General Motors, and Honda, but the imported models produced or introduced in China have almost no outstanding performance in high-end intelligent driving.
If you look at the industry level in 2023, their performance is failing. Moreover, in the foreseeable next two years or so, their intelligent driving ability is likely to be able to change lanes with lights on, and the possibility of landing high-speed navigation assistance is relatively low. The only thing that is more interesting among the joint venture car companies should be the core Cheng of the joint venture between Volkswagen and Horizon. 
There is also a certain gap between the intelligent driving level of traditional independent brands and new forces. If the general is plucked among the dwarf, Geely is the one with a relatively small gap.
Geely can be said to be full of flowers in high-end intelligent driving, and it is precisely because of the blessing of self-development, first-class business solutions, cooperation and other ways that Geely's various brands have not been too lame in terms of intelligent driving.
Although ZEEKR 001 has been delivered for more than 2 years, the previously advertised intelligent driving NZP has not been opened, how to make these two world's first Mobileye Q5H chips play their capabilities, has become a big problem, and the same is true for ZEEKR 009. Fortunately, we saw on the latest ZEEKR 007 that ZEEKR has begun to use a new solution based on NVIDIA's ORIN chip, which has finally narrowed the gap between ZEEKR and the new forces in terms of intelligent driving.
Overall, Geely has made some progress in cooperation and self-development. For example, Geely's Boyue L model launched in 2022, a model with a configuration of about 150,000 yuan, has a high-speed pilot assisted driving function based on the Horizon Journey 3 chip, which lowers the threshold for high-end intelligent driving. ECARX's self-developed chips also give more possibilities for the next competition. 
Compared with Geely, other independent brands have fewer functions in terms of intelligent driving.
At the Star Era ES released at the end of 2023, Chery will take LiDAR + NVIDIA Orin chip for the first time to move towards high-end intelligent driving. Some of the Wei brand models under the Great Wall, although they have been equipped with LiDAR + Qualcomm's intelligent driving chip early, are not ideal in terms of landing effect. Of course, this has a lot to do with the sales of Wei brand models.
As for BYD, everyone has always said that its intelligent performance is mediocre, and it was indeed the case before. However, the new Yangwang U8 and Denza N7 models have already started to be equipped with LiDAR. It's just that BYD didn't vigorously promote these configurations and highlights, and there are not many functions that can be seen on the ground. 
The high-precision map has fallen out of favor, and the lidar is not fragrantAlthough the capabilities of various car companies in intelligent driving are uneven, this gap will be narrowed by 2024.
In fact, from 2022 to 2023, the development rate of the field of intelligent driving has slowed down somewhat. On the one hand, the impact of the winter of autonomous driving has not passed, and due to the upper limit of autonomous driving, the industry expectation has hit a bottleneck; On the other hand, the openness of policies and regulations also affects the progress of assisted driving to autonomous driving.
In addition to the influence of the above two industries, there is also a technical reason, that is, the underlying technology development route of intelligent driving is becoming more and more unified. This leads to another great change in the industry in 2023: high-precision maps and lidars are gradually abandoned, and pure vision + large model algorithms are increasingly strengthened. 
Starting in the first half of 2023, Wei Xiaoli and Huawei have invariably switched to the route of BEV+ spatial network technology, and almost simultaneously announced that they will gradually no longer rely on high-definition maps. By the end of 2023, they have indeed weakened their reliance on HD maps. The reason is very simple, if you want to achieve high-end intelligent driving in all scenarios, you must cover places where there is no high-precision map, so you must strengthen the vehicle's own perception and decision-making ability.
Two years ago, the industry proved to us that as long as there is a high-definition map, the new forces can give an "autopilot"-like experience. Without considering other factors, with the blessing of high-precision maps, not only will car companies reduce development costs, but also the pressure on the computing platform will be reduced, and the perception of vehicles will become more accurate.
However, the high cost, low freshness, and low regional coverage of high-precision maps will not be fundamentally changed for a long time in the future. Car companies want to do high-end intelligent assisted driving that "can be used and is easy to drive", and high-precision maps really can't be counted on. Therefore, it is necessary to strengthen the perception ability of the bicycle and the computing power of the background cloud. 
Tesla released and used BEV and occupancy network technologies in 2021 and 2022 respectively, and we will also see these two technologies in the products of new forces in 2023. Coupled with the comprehensive explosion of artificial intelligence technology last year, the route of the intelligent driving industry has been almost unified as "BEV + Transformer + Occupy Network", and the ability of visual algorithms has been continuously strengthened.
In addition to high-precision maps, we found that lidar, which was a hot topic in the industry a year ago, was also somewhat injured, and its presence gradually decreased. Nowadays, there is a pure visual intelligent driving system in mass production in China.
2022 is known as the first year of LiDAR on the car, and models equipped with LiDAR such as Xpeng P5, Xpeng G9, NIO ET7, Ideal L9, and AVATR 11 will be launched and delivered in large quantities, and the multi-LiDAR solution is even more popular. There is even a player like the Salon Mecha Dragon, which is equipped with 4 lidars, which gave birth to the classic slogan: Don't speak if you are less than 4.
But in 2023, a large number of single lidars, or even models without lidar, will become mainstream players. Throughout last year, new models equipped with multiple lidars can be counted almost on one hand, such as Xpeng P7i, Xpeng G6, AVATR 12, Denza N7, Yangwang U8, and a NIO ET9 that will not be delivered until 2025. 
The multi-lidar approach was very popular in 2020-2022, and there was a perception at the time that it was more gimmicky than practical. However, at that time, we had nothing to compare, and the high-end intelligent driving solutions of each company were still in the PPT stage. Until the ARCFOX Alpha S Huawei HI version model equipped with 3 Huawei lidars came out, we saw the strength of the multi-lidar solution. The AVATR 11 and Salon Mecha Dragon that appeared later were also equipped with Huawei's multiple lidar solutions.
But in 2023, the situation has changed abruptly. Huawei released the M5 intelligent driving version in April, equipped with the latest Huawei ADS 20 system, only 1 128-line lidar in terms of hardware, as well as a forward-looking binocular camera, the core computing unit uses MDC610 of a single board, and the computing power is 200tops.
AVATR 11 uses three Huawei-equivalent 96-line semi-solid-state lidars, as well as front-looking quadocular cameras, and the core computing unit is MDC810 with a single board and a computing power of 400TOPS.
Through a simple comparison, you can see the gap between the software and hardware of the two systems. However, the M5 ADS 2There is not much difference between 0 and AVATR 11 in terms of functional performance. It is equivalent to Huawei making a set of ADS 2 that distinguishes between high and low configuration hardware0, but it doesn't distinguish between functional differences, and it feels a little strange no matter how you think about it. 
The reason behind this is not difficult to understand, that is, to reduce costs. Especially in the context of the first war, most of the models can't afford to use the top scheme of 3 lidars. Therefore, Huawei has to cater to the market to do subtraction.
Huawei's approach is actually a microcosm of market changes. We're seeing similar initiatives from Horizon, Mobileye, and even Qualcomm. In the end, everyone is moving towards the pure visual direction of medium and low computing power, and lidar is destined to be abandoned.
Here you may have questions, why the sales of domestic lidar manufacturers are getting better and better, and there is no sign of decay at all. We think that in 2024, it may be difficult to see such beautiful data again.
At present, we have collected the shipment data of the leading lidar manufacturers Hesai Technology and Suteng Juchuang
In the first three quarters of 2023, Hesai Technology's total deliveries exceeded 1340,000 units, a year-on-year increase of 3079%；Among them, the delivery volume of vehicle-mounted lidar exceeded 1140,000 units, a year-on-year increase of 5166%。Suteng Juchuang's LiDAR delivery volume reached 1060,000 units, of which the total sales volume of vehicle-mounted lidar exceeds 930,000 units, sales reached a new high.
This kind of data performance is actually not difficult to interpret, Hesai Technology's AT128 LiDAR is currently the biggest customer is Ideal, and in 2023, the sales of Ideal will explode, so it is easy to understand that Hesai Technology has such a result.
In 2024, Hesai's delivery volume is likely to continue to grow, but the supply of ideal supply will definitely slow down or even decrease. Because the two LiDAR-free versions of the Air and Pro have become the main sales models of the ideal L-series models.
Next, Hesai's new growth point may come from Xiaomi cars, of course, it also depends on the pricing of Xiaomi SU7. In the future, the growth point of Hesai Technology is still unclear.
The same is true for Suteng Juchuang, which has more customers than Hesai AT128, so it is natural to have good results. Suteng Juchuang's good results may last longer, but its continued huge losses are no less than some new head car manufacturers. Insufficient production capacity, low gross profit margin, and loss of orders are the big problems that Suteng Juchuang has to face.
The madness of lidar companies may be punctured at any time, on the one hand, from internal forces, that is, the new forces are gradually moving towards pure vision and diluting lidar. On the other hand, it comes from an external force, that is, Tesla's FSD, which may attack at any time.
As we mentioned earlier, the intelligent driving technology used by the new forces will more or less "pay tribute" to Tesla FSD, but the full version of Tesla FSD is currently only limited to North America. With the new generation of FSD chips and the latest vision algorithms landing on the new Model S X, as well as the facelift of Tesla's Model 3, Tesla's new generation of intelligent driving will be quickly rolled out.
At the same time, Musk also announced that he would open source all the core technologies of Tesla's automatic driving, and in the future, anyone may be able to browse the core technology of Tesla's FSD on the Internet. This is definitely a sword of Damocles hanging over their heads for the new forces who are bent on self-development.
End-to-end FSD, which has low hardware cost and high efficiency, is almost a dimensionality reduction blow to the entire industry. After rolling for a long time, everyone is likely to be "bloodbathed" by a FSD that is not as expensive as 1 lidar with all the intelligent driving hardware added together. As the research and development and landing in the field of intelligent driving are slowing down, if this "catfish" is put in, the big wave will be set off, not only for the new power car companies engaged in self-development, but also for the best businessmen.
Calm down, L3 isn't so godlyThe centralized issuance of L3 test licenses at the end of 2023 can be said to be a shot in the arm for the whole industry, and traditional car companies account for the vast majority of them. Mercedes-Benz and BMW, in particular, seem to be muffling to do big things. But what is the essence of the L3 they made different from the L2+++ made by the new forces?
For the current regulations and policies, these L3 conditional autonomous driving are actually of little significance, and they are no different from L2. Since there is not much formal legal reference, we still refer to the L3 measures previously promulgated by Shenzhen, which clearly mentions the division of responsibilities. In the event of an accident with an autonomous driving vehicle, the driver is still the first party responsible, and if there is evidence that it is a vehicle system problem, then the car company or developer can be held responsible. But how can this evidence be so easy to find, so in this way, the division of responsibility for L3 is almost the same as that of L2.
The L3 is conditional autonomous driving, and we can also see the L3 that Mercedes-Benz has landed in Germany and the United States, and the automatic driving can only be turned on at a speed of less than 60km h in high-speed scenarios. For Germany, which has a large number of unrestricted speeds, this L3 is of little significance, and even dangerous. If this kind of L3 also comes to China, it seems that it is of little significance at present.
On the other hand, the so-called L2+ that domestic car companies are rolling up is actually different from the traditional L2. The L2 solution, which is commonly used by joint ventures, only assists in the longitudinal aspect, that is, it is responsible for adjusting the accelerator and brake and vehicle distance.
The L2+ done by independent brands and new forces adds horizontal lane changes, entry and exit ramps and other functions, which is actually what L3 autonomous driving should do. This makes the Mercedes-Benz BMW L3 look even less bright.
Of course, since Level 3 is called autonomous driving, it means that it is still different from Level 2. The domestic L3 can move from a specific test path to a scene closer to our lives, and this process will be accelerated.
Write at the endLooking back on the intelligent driving industry in 2023, you will find that there is not much difference between the new forces' problem-solving ideas and the final actual performance after they collectively submit the papers. Even 100,000 yuan of intelligent driving. It is no different from the intelligent driving ability of million-level models.
There is also an interesting point, that is, the sooner or later of entering the game does not seem to have a direct relationship with the final ability performance of intelligent driving.
Xiaomi's entry into the game is very late, but it has caught up with the critical opportunity for the whole industry to switch its technical route. Therefore, Xiaomi's intelligent driving has taken a lot of detours. Coupled with the explosion of artificial intelligence large model technology, Xiaomi's mass production intelligent driving ability can almost directly enter the first echelon.
In addition, the influence of factors such as the industry's best war has led to the high-cost hardware of lidar, which is no longer so favored, and when Tesla really comes in, a new reshuffle is about to begin, and the big drama in 2024 has just begun.