Jia Haonan sent from the co-driving temple smart car reference | Official account AI4Auto
"LiDAR is life-saving, nothing is more important than life safety, and there is no point in shouting low prices to earn eyeballs."
When everyone was eagerly looking forward to a price that was more gimmicky than the "thousand-yuan machine", CEO Li Yifan came directly.
Hesai doesn't roll the price of anything, can it win this group of menacing "thousand-yuan machines"?
And we also have to see how Hesai's new lidar can "save lives".
What kind of new products did Hesai release?
Hesai's new LiDAR ATX:
Judging from the name, it is a platform-based product, which is the same source as the AT series that has been mass-produced for hundreds of thousands of units before.
Equipped with Hesai's fourth-generation chip architecture, the "new" is visibly improved to the naked eye, with a 60% smaller volume than the AT128 and a minimum exposed window height of only 25 mm:
The weight is reduced by 50% (only 500g) while the power consumption of the whole machine is reduced to only 8 W.
The ATX has a maximum detection range of 300 meters, which is 50% higher than the AT128. It supports up to 256 lines with an optimal angular resolution of 0.08° x 0.1°, which is more than twice that of the AT128.
In horizontal comparison, the main vehicle specification front-view lidar products on the market generally have a detection distance of 150-250m, and the angular resolution is generally about 0.2° x 0.2°.
As you can see, Hesai emphasizes the highest number of lines, rather than the common "equivalent number of lines".
In fact, the equivalent number of lines is commonly found in MEMS and two-dimensional mirror technology architectures, which often have only one or several lasers, and then cooperate with a scanning mechanism to achieve a fixed range of scanning, which is equivalent to "the same effect as scanning with a mechanical/one-dimensional mirror lidar with XXX lines".
However, due to the limited operating frequency of the rotating mirror or MEMS galvanometer, the overall point cloud density/exit velocity is still inferior to the traditional mechanical radar and one-dimensional rotating mirror radar with high indicators.
Recently, some manufacturers have shouted a new term of "ROI region equivalence", ROI is the encryption area, which refers to a small area in the middle of the lidar field of view where the point cloud is denser. This kind of "ROI equivalence" statement is quite a propaganda clever component. Hesai CEO Li Yifan has a very vivid analogy:
This is like a person who received a year-end bonus of 100,000 yuan one day, and then told others that "the equivalent annualized bonus is 30 million".
Hesai's automotive-grade mass-produced products use a one-dimensional rotating mirror architecture, and the performance improvement depends on the degree of chipization of the transceiver module. Hesai also confirmed that the maximum number of ATX supported lines of 256 is the real number of lines, which is not equivalent:
The main advantages of the ATX are that the thickness is only 25mm, and the installation position is flexible, which greatly improves the aesthetics of the vehicle. Can be placed on the roof of the car:
Can be installed behind the windshield:
It can even be integrated into the headlight assembly:
Secondly, Hesai's optical architecture design in the ATX enables the ATX to have a 7x optical zoom function. When driving in urban areas, the ATX can achieve an ultra-wide field of view of 140° to adapt to the complex road conditions in urban areas, and to maximize the monitoring of road conditions on both sides of the vehicle.
At high speeds, the 7x optical zoom enables distance measurement at a distance of up to 500 meters and 7x HD perception:
By the way, the Hesai ATX 140° ultra-wide field of view also surpasses the common 120° products:
A wider viewing angle is one thing, and on the other hand, it can also save part of the cost of blind radar.
To sum up, Hesai's new ATX is not a "just enough" pragmatism, and it can be called the "king" of this year's lidar in terms of detection range, resolution, size, etc.
Li Yifan: LiDAR saves lives, and the low price is meaningless
Suddenly, the laser mine belt entered the era of "thousand-yuan machine".
In the previous few important releases, almost no one cared about the parameters and performance, and shouting a low price became the most important gimmick.
So how much can ATX shout for?
Hesai CEO Li Yifan has the following view:
Lidar is like an airbag, it can save lives at critical moments, and it makes no sense to lower the price and shout to earn eyeballs.
This kind of publicity will make the public ignore and misunderstand the importance of lidar in the intelligent driving system.
The mass-produced intelligent driving system can be seen as an AI driver, the same as humans, and also uses the eyes (camera) to perceive all information. The "brain" of the AI driver is the perception and recognition algorithm behind it.
The essence of the algorithm is AI—a deep learning algorithm, not an "if else" judgment instruction, because it is impossible to exhaustively enumerate every traffic participant or scenario situation in the world. Since it is AI, it requires a large amount of data learning as the basis for perception and recognition.
However, there is a problem here, there are endless aliens and special-shaped cars on the road, and it is difficult for AI to learn and train on a limited dataset to fully recognize and cover.
Therefore, adding a sense to the pure visual system with bad "eyes" has become the best way - LiDAR comes in.
Lidar emits infrared light, and when it encounters obstacles, it will generate corresponding echo signals, which naturally include the shape and size of the target, and through the echo time, the distance information can also be calculated.
It is equivalent to the lidar through the "sense of touch", to build the 3-dimensional data of the scene, to the system as a reference supplement to the image data: if the visual error detection is missed, the lidar tells you that there is a XX target.
Therefore, LiDAR is the perfect redundancy and back-up means for the intelligent driving system.
Taking AEB (Active Braking) as an example, LiDAR with better performance represents stronger ranging capabilities, which directly affects the upper limit of AEB's effective speed:
Therefore, no matter what stage the smart car industry develops, the performance of lidar will always be the first element that car companies and users care about the most.
For example, ATX's fourth-generation chip architecture uses 3D stacking technology to integrate 512 channels on a single board. 8-core APU with 24.6 billion samples per second. The detector sensitivity has been increased by 130%, and the power consumption of single-point ranging has been reduced by 85%.
In addition, Hesai has developed its own LiDAR "discrete graphics card", the Intelligent Point Cloud Engine (IPE), which integrates a 256-core waveform processing core to achieve fine capture of the full waveform with a high sampling rate of 24.6 billion times per second.
The main function is to allow the lidar to intelligently identify rain and fog, front vehicle exhaust, and dust, filter out more than 99.9% of the environmental noise, and greatly reduce the interference of rain and fog weather on the lidar.
Hesai officially announced that the ATX will be launched in 2025, and has already obtained mass production designation for more than 10 models:
2024, LiDAR "Rapid Convergence"
Last year was the "first year of LiDAR", and a large number of models equipped with LiDAR were delivered, and ordinary users began to recognize this new sensor.
In the first year, Hesai ran out of the pattern of "taking the lead":
△ Source: Public data on the amount of insurance on new vehicles
What kind of lidar manufacturers will car companies choose?
Judging from the several leading players that have come out, it can be summarized into these three indicators: product performance, engineering ability, and business status.
They correspond to the three stages of LiDAR since its birth: the sample stage, the product stage, and the commodity stage.
The sample stage refers to the era when lidar sensors were just being born and startups were entering the game. The landmark event was the launch of LiDAR at the North American DARPA Challenge, giving birth to the entire autonomous driving industry.
At this stage, LiDAR competes for technical routes, such as rotating mirrors, galvanometers, prisms, etc., and each scheme can cite a lot of advantages. To put it bluntly, this stage is the To VC stage of LiDAR.
The product stage is from the beginning of Robotaxi to the rise of mass production of intelligent driving, and the stage of LiDAR from vehicle to vehicle specification, until 2024, it is the product stage.
At this stage, the development of lidar is clearly centered on the needs of car companies, and the core of competition is the engineering ability and stable delivery manufacturing capacity that meets vehicle regulations.
The first pioneers to be eliminated also appeared: Ibeo, Velodyne...
In these two stages, Hesai made some "anti-consensus" choices at that time.
For example, in terms of technical routes, Hesai insisted on 905nm near-infrared light, rather than the 1550nm far-infrared route, which was more favored at the time.
In the early stage, there was also a dispute between the route of chips and discrete components, as well as the different propositions of foundry and self-built factories.
Hesai's choices along the way, in retrospect, have become the basis for nearly half of the ADAS market share today, as well as the advantages favored by most car companies.
For example, practice has proved that 1550nm infrared light has become the biggest problem due to the special requirements of lasers for materials, and it has no chance to popularize intelligent driving.
In terms of LiDAR architecture, Hesai has always adopted a one-dimensional rotating mirror architecture, which has the advantages of being explained earlier, with a reliable structure and stable performance output. In the past two years, it can also be observed that most players have switched to 905nm and the one-dimensional rotating mirror route.
Chip-based and self-built factories correspond to the capabilities of LiDAR players in terms of performance-cost balance and project delivery, respectively.
Hesai's script from the beginning to the present can be applied to almost any player who successfully survived the first two stages of competition.
This also directly illustrates a problem: in the lidar industry, the choice of technical route is not the key, several architectural principles are not confidential, the threshold is not high, and the switching cost is affordable, so as not to hurt the muscles and bones.
So in 2024, when we enter the commodity stage of lidar, what will we compete for?
On the surface, it seems to be a "low volume".
Hesai does not deny the trend of thousand-yuan machines. However, Li Yifan emphasized that the "thousand-yuan machine" is the price, not the cost.
The industry has its own development cycle, and the consideration of profits by car companies directly affects the purchase price of lidar. Players who have the strength to survive until now have no essential difference in the terminal price.
In other words, if LiDAR players want to stay at the table this year, the offer cannot deviate too much from the "thousand-dollar machine". The same is certainly true of Hesai.
Therefore, what really determines life and death is the cost control ability behind the "thousand yuan" selling price.
Simplified into financial indicators, it is actually gross profit - selling one and losing one, or healthy and positive self-hematopoiesis. After all, car companies certainly don't want their core suppliers to be precarious.
Taking Hesai as an example, the comprehensive gross profit margin in 2023 will be 35%, ranking first in the industry. In addition, from the perspective of the "blood bar" (cash/quarterly operating cash flow) of the company's operation, Hesai is also the only player who has "no worries about life":
In detail, Hesai's high gross profit is the result of several previous choices: cost reduction and efficiency increase at the manufacturing level, LiDAR chipization at the technical level, and independent integration of the supply chain.
If 2023 is the first year of getting on the car, 2024 is the "first year of the explosion" of lidar.
Reflected in two data, the first is the penetration rate, Hesai cites data from the Gaogong Intelligent Vehicle Research Institute to believe that the proportion of lidar models will reach 16% in 2024 - the so-called key node of "crossing the chasm".
Secondly, Moore's Law of LiDAR is still in effect, and the core of crossing the 16% node is that under the premise of maintaining a certain performance, the cost of LiDAR can be standardized to 150,000 models - the annual market stock exceeds 6 million.
However, this year's knockout indicators are also stricter than in previous years.
In addition to engineering capabilities and mass production delivery capabilities, it is also necessary to add the health of operating cash flow, and the basis of this indicator involves the degree of chipization, mass production scale, and self-sufficiency of key components...
Li Yifan, CEO of Hesai, judged that 2024 has not yet run out, and the window of opportunity is not absent, but it is getting smaller and smaller.