Shentucar original
Author | Zhou Jifeng
Edit | dawn
In 2021, whether it is a veteran of the car industry, a new egg, or a PPT fool, invariably launched a lidar model.
Lidar, what Musk called "only a fool would use", became the absolute "C bit" in the car-making industry.
You don't know what kind of "goods" it is, but it is on fire, and this fire is burning more and more vigorously. Even in new cars, they are embarrassed to show their automatic driving ability without lidar.
Cars equipped with lidar are not cheap, and Guojin Securities pointed out that it has confirmed that the production of front-loading mass production models will exceed 300,000 units in 2023, and the price is concentrated in 400,000-800,000 yuan. The wool came out of the sheep. Will this wave of lidar fever be a new wave of leek party revelry?
If you want L4 to get by, lidar can't be less?
How hot is lidar, the following table is very illustrative.
The deep way is not completely statistical
In 2022, there will be as many as 12 new energy vehicles equipped with lidar. Almost all known electric vehicle brands have joined the arms race. Beiqi Jihu Alpha S, Xiaopeng P5, Weilai ET7, SAIC RS33, Zhiji L7 and other new cars have all previewed the launch of lidar.
1 lidar is not even enough, the head of the salon mecha of the New Energy Vehicle Brand under the Great Wall shouted at the Guangzhou Auto Show: "How many lidar do you have?" If you have less than 4 pieces, please do not talk. "There are also players who incarnate as ruthless lidar wholesalers, with 5 lidar on a car.
Such a crazy arms race is based on an increasingly fashionable consensus: if L4 is to be achieved, lidar cannot be less.
Autonomous driving has three major elements: "perception, decision-making, and control." Each link needs specific hardware and software to support. At the perceptual level, electric vehicles need to rely on sensors to perceive and transmit external information. These sensors are the equivalent of the "eyes" of a car.
At present, the sensors on electric vehicles include: vision sensors, millimeter wave radar, ultrasonic radar, lidar and on-board sensor networks.
Different types of "eye" probes are also detected at different distances and in precision.
The car has to go on the road, it has to face extremely complex reality scenarios, and to achieve a higher level of automatic driving capabilities, it also puts forward higher requirements for the "eyes".
Millimeter-wave radar is low in cost, but the detection distance and accuracy are relatively weak, and the vision sensor is easily affected by the weather. Car companies have been wandering around like headless flies for many years, and then they have gradually found a relatively feasible path at present - the lidar route.
The high accuracy and long detection distance of lidar can enhance the redundancy of the perception system and supplement the missing scenes of millimeter-wave radar and cameras. Aside from being susceptible to natural light or thermal radiation, and being expensive, lidar has few drawbacks.
Moreover, since entering 2020, the cost of lidar has been relatively less expensive. Semi-solid-state, solid-state lidar gradually replaced mechanical radar, and the cost of lidar was reduced from tens of thousands of dollars to $1,000. Lidar is also shrinking in size, evolving from a stupid and big "black guy" to a "part" that can be hidden in the car body.
The cost is low, the technology has advanced, and the consensus has been formed, which has led to a large number of lidar on the car.
Since entering 2021, the news of lidar manufacturers' financing has continued.
In August, lidar maker Innocent announced a $66 million Series B+ round of funding. Its investor list includes NIO. Nio's new car ET7 is equipped with ultra-long-range high-precision lidar provided by Innocent. In November, LiDAR manufacturer Hesai Technology announced that it had received additional financing of 70 million US dollars from Xiaomi Production and Investment, and with the previously announced financing of more than 300 million US dollars, the total financing of Hesai Series D has exceeded 370 million US dollars.
Traditional automakers such as Ford, Volvo, Daimler, and Mercedes-Benz are actively involved in the lidar field through investment or cooperation.
The more lidar, the more capable?
The melon-eating masses may wonder why the lidar on the car is not installed in a fixed position, for example, the lidar of the Weilai ET7 is overhead, while the lidar of the Xiaopeng P5 is on both sides of the front bumper. The number of lidar carried by car companies is also not uniform, giving people a feeling that the more lidar, the stronger the ability to automatically drive.
NIO ET7, lidar overhead
Let's first talk about how lidar works.
Lidar can be divided into four major systems: laser transmission, laser reception, beam manipulation and information processing. Lidar emits a beam of light first, the beam reflects when it encounters an obstacle, and the optical receiver receives a pulse of light reflected back by the object. Subsequently, the information processor on the lidar will map the three-dimensional point cloud of the surrounding objects by measuring the time difference and phase difference of the laser signal, thereby obtaining accurate distance and contour information.
But how far and accurate a lidar can measure, the water inside can be deep.
Lidar has different technical routes in five aspects: ranging principle, laser transmission, laser reception, beam manipulation and information processing. Manufacturers take different routes, and the performance and effect of the lidar created will be very different.
There are also multiple dimensions for measuring the performance of a lidar on the market, including the number of lines, detection distance, measurement accuracy, measurement accuracy, scanning frequency, and so on.
There is no uniform evaluation standard in the industry. Everyone was crossing the river by feeling the stones. No one seems to be able to say how many lidars a car should carry, what route the lidar should take, and how much the price should be set.
Therefore, although they are all lidar, the location of different lidar installations is not uniform, the cost is not the same, and the performance and effect are also very different.
At present, the rare consensus is that everyone is going in the direction of pure solid-state lidar.
The most expensive hardware unit in lidar is the scanning component, which directly affects the performance of lidar. According to the movement mode of the scanning module and the transceiver module, lidar is divided into three categories:
Mechanical: The lidar rotates as a whole to scan the object 360 degrees. High cost, large size, mature technology, mostly used in Robotaxi.
Hybrid solid-state: The laser transceiver module is not in motion, only the scanning module is in motion. It is further subdivided into three routes: MEMS, rotary mirror type and prism type.
Pure solid state: Not only does the laser transceiver module not move, but the scanning module also has no mechanical motion. This route is divided into OPA and Flash.
Mechanical lidar technology is mature but the cost is too high, the best solution is a pure solid-state solution, solid-state lidar has no moving parts, so the size is smaller, lower cost, and the reliability is higher, but the technology of pure solid-state solution is not yet mature. Therefore, most models choose to be equipped with semi-solid-state lidar, and wait until the technology is mature to advance to pure solid-state lidar.
For example, the lidar equipped with the NIO ET7 belongs to the rotary mirror lidar. This type of lidar has the advantages of long detection distance and large detection angle. In order to better scan objects, car companies generally place this lidar on the roof.
There are also bold people, such as the Great Wall Mocha, who have chosen to carry solid-state lidar directly.
Is it okay not to engage in lidar?
So the question is, is it okay not to carry lidar?
OK.
Tesla is adamantly opposed to lidar.
At present, in the car-making jianghu, it is divided into two factions, one faction takes the pure visual route, the sensor is mainly camera + millimeter wave radar, focusing on artificial intelligence visual algorithms, this faction has and only one car company is insisting, that is, Tesla. The other faction insists on taking the lidar route, the sensors on the car are mainly lidar, supplemented by millimeter-wave radar, cameras, etc., and most car companies take this route.
Musk is the absolute opinion leader in the car industry. However, in the choice of perception layer technology route, most car companies are on the opposite side of Musk.
It's not that other players don't want to follow Musk, it's that most car companies can't keep up.
The essence of autonomous driving is to rely on a large amount of data to feed and train AI. Tesla has super algorithms and road test data that are difficult for others to match. In other words, the "brain" of the car is smart and nutritious.
Tesla electric vehicles all over the world are collecting data for Tesla. By the end of 2016, Tesla had 3.5 billion miles of road test data. This vast amount of data is constantly feeding Tesla's AI brain.
Source / pexels
In the core AI algorithm, Tesla has been far ahead of other car companies. Tesla has been developing its own self-driving chip since 2019, and its new generation of chips, HW4.0, has gone into testing and is said to have 3 times the performance of the old version. In addition, Tesla also has AI supercomputer Dojo. The Dojo project leader said tesla Dojo is the fastest AI training computer in history.
In terms of algorithms and road test data, other players are inferior to Tesla, and taking the lidar route is the optimal solution.
The environment of the inner volume also makes the car companies involuntarily.
"Now the entire industry has basically formed a consensus, except for Tesla, as long as it is an electric vehicle, it must be equipped with lidar." Analysys analyst He Qi pointed out to Shentu.
In this carnival in the name of technology, the most active ones are the new car-making forces and independent brands. These players are particularly aware that if they do not seize this opportunity, the future wave of car manufacturing will sweep the world, and they will eventually be thrown off the car.
Roland Berger Global Partner Johan Karlberg believes that profits from the automotive industry will be redistributed based on autonomous driving. On the road through high-level autonomous driving, lidar is currently the best paving stone.
Many players have seized the commanding heights of technology by embedding lidar and rolling up the number of hardware, which is equivalent to getting a ticket to the future.
Software capabilities are not enough, hardware to make up
2021 is called the first year of lidar, but lidar still has a long and tortuous road to go.
In addition to technology research and development, the biggest bug of lidar is that the cost is still very high. Everbright Securities pointed out in the research report that most of the excellent performance of lidar prices around $1,000. The cost of lidar still accounts for the majority of the comprehensive cost of smart auto parts.
Buyers want the price of lidar in L2 to drop to $500 and the price in high-end autonomous driving systems to $1,000.
There is still a large gap between ideals and reality.
"Lidar is the need to go volume, through the pull of the market scale to promote price exploration." Now ultrasonic radar or millimeter-wave radar is more mature, and it can be bought for hundreds of dollars. He Qi's analysis pointed out: "Under the continuous market education, lidar may be like millimeter-wave radar, forming a certain market scale, low cost, and most people can afford it." ”
Large-scale mass production can share high R&D costs and production costs. Ruichi Zhiguang disclosed that the current supply price of LakiBeam128 is set at $743; at the level of 100,000 units, its unit price is $498; at the level of one million units, its unit price is $289, and the price brought about by mass production is significant.
"Overall, lidar is not a capital bubble, but a proven solution with a certain market demand. The crazy inner volume of the entire industry is actually seeing that this road can be done. Ho Chi said.
Of course, hardware stacking is only a precursor to autonomous driving. Lidar is equivalent to the eyes of the car, and the level of car intelligence is determined by the "brain", that is, AI algorithms and software capabilities.
Nowadays, most of the domestic car companies are still fighting hardware, "stacking materials", software systems can only rely on external forces, and only a very small number of players have decided to develop full stacks.
From time to time, there are opponents on the Internet who complain: "Domestic smart cars will only 'pile up'." The 'soul' cannot keep up with the 'body'. In a way, that's true.
Car companies are playing such a calculation: the hardware of self-driving cars determines the upper limit of automatic driving capabilities, and now, it is not yet time for software capabilities to compete, but the hardware has been available, so that they can get tickets for the next round of competition.
"I think there are pros and cons to 'stacking', the industry uses a lot of lidar, and when it is measured, the cost will fall, which is a good thing." He Qi said: "Of course, for car companies, if the software capabilities are not enough, the speed of system update iteration will be very slow. However, this is also a way, like everyone is climbing a hill, if I can't climb up, then why not choose to use external forces? ”
*The caption image is from Visual China.