Pengpai News reporter Shao Wen
2022 at the beginning of the year, on January 3, US stock lidar concept stock Weishi Image (MVIS.US) rose 8.38%, Luminar (LAZR.US) rose more than 5%, and Velodyne (VLDR.US) rose 4.95%.
Guojin Securities predicts in the report "The volume is imminent, the first year of lidar installation before it starts", predicts that lidar is expected to rapidly reduce its price through large-scale mass production + technological progress, exceeding US$23 billion in 2030, with a compound growth rate of nearly 90% from 2021 to 2030, and the overall penetration rate of front-installation reaches 45%.
In 2020, the industry leading companies Velodyne, Luminar and others have been launched one after another. The announced plans to mass-produce lidar models in 2022 have reached 15. Car companies such as Xiaopeng , NIO , SAIC , Mercedes-Benz and other car companies have successively made previews on new lidar models.
vehicle manufacturer has announced the price range and output forecast of mass-produced models equipped with lidar (Taiwan). Source: Guojin Securities Research Institute
Global Market Research and Strategy Consulting Company Yole Development also predicted in the "2021 Lidar Application Report on Automotive and Industrial Fields" that ADAS (Advanced Driver Assistance Systems, Advanced Driver Assistance Systems) and robot vehicles will drive the radar market to grow from US$1.8 billion in 2020 to US$5.7 billion in 2026.
"Looking at the world, all lidar manufacturers are still in the rising stage. It is expected that around 2035, the lidar market for vehicles will reach the range of US$13 billion to US$13.5 billion," Sinclair Vass, chief commercial officer of Velodyne, said in an interview with Pengpai News (www.thepaper.cn).
At present, lidar has not yet reached the stage of mass production, and many technical routes are still being explored and are in a state of blooming flowers. What are the current trends in the technological development of lidar? Will the "visual school" and "lidar school" that are divided on the perspective of autonomous driving perception develop in the future? Different technical routes of
lidar: mechanical and solid-state
lidar was first invented in the 1960s. It was mainly used in space exploration, terrain surveying, weapon guidance, etc. In 1971, lidar completed lunar surveying and mapping with Apollo 15 . In 2005, Velodyne equipped self-driving vehicles with lidar for the first time at the DARPA challenge in the United States. In 2007, Velodyne produced the first commercial 3D dynamic scanning lidar, and the commercialization of lidar began. At present, lidar is widely used in autonomous driving, logistics and transportation, high-precision maps, smart transportation, robots, industrial automation, drones, surveying and mapping and other fields.
Source: Autobot Reference, Guojin Securities Research Institute
LiDAR uses visible light and near-infrared light to emit a signal, and is reflected by the target and is collected by the receiving system. The target distance is determined by measuring the running time of the reflected light. According to Guojin Securities' report "The volume is imminent, the first year of the lidar installation before it is launched", shorter wavelength and active laser technology empower the advantages of lidar with high measurement resolution, long detection distance, large detection angle, strong night working ability, and strong anti-interference ability, which can directly obtain information such as distance, angle, reflection intensity, speed, etc.
LiDAR main explicit parameters Source: Hesai Technology Prospectus
LiDAR is often distinguished by the number of lines, which refers to the number of light sources emitted by the laser. For example, 128 line products have 128 light sources. According to the scanning method, whether there are mechanical rotating parts in the laser radar scanning method can be divided into mechanical, semi-solid state, and solid state. Mixed solid state is divided into MEMS and rotary mirrors, and pure solid state is divided into phased array OPA and Flash. Lidar has different technical routes in five aspects: ranging principle, laser emission, laser reception, beam manipulation and information processing.
Source: Guojin Securities Research Institute
"Different architectures, such as mechanical, MEMS and solid state, have different advantages and disadvantages for different applications. However, what is actually extremely important is the specific optical design performance achieved, the quality of production and large-scale feasibility, and the ability to achieve attractive price points. In addition, the ability to combine lidar hardware with high-performance perception software such as the Vella platform is a major differentiator," said Sinclair Vass.
Currently, there is a clear trend in the field of lidar. Mechanical lidar with mature technology, excellent detection performance, high cost, low life, and large volumes are evolving towards solid-state lidar.
According to the "2021 LiDAR Application Report on Automotive and Industrial Fields", after 2020, the mechanical (including rotary, mirror-type, etc.) lidar technology routes will gradually decline. Due to the maturity and cost advantages of upstream components, the MEMS and Flash technology routes will gradually become mainstream.
mechanical type refers to the arrangement of multi-beam lasers in the vertical direction, and the photoelectric structure is driven by a motor to rotate by 360°, thereby turning points into lines to form a three-dimensional point cloud solution. Its line number is proportional to the resolution, has the characteristics of high resolution and high ranging. It is the most mature solution at present and the highest output. However, its cost is high. At the same time, in order to achieve high-frequency accurate rotation, its mechanical structure is complex, with an average failure time of only 1000-3000 hours, which is a significant gap from the minimum 13000 hours required by the automotive specifications, making it difficult to achieve mass production of pre-installation.
LiDAR pricing Source: Industry research, Guojin Securities Research Institute
solid state refers to the type of lidar without any mechanical moving parts. It can be subdivided into OPA, Flash, electronic scanning and other forms. The current technology is relatively mature. Taking OPA (optical phased array technology) as an example, OPA uses voltage regulation to achieve beam deflection to realize beam deflection, which has the advantages of fast scanning, high accuracy, small size, strong controllability, and strong vibration resistance. After the technological breakthrough, the cost is low and the degree of mass production standardization is high. It is considered by some industry experts to be the ultimate mainstream form of lidar.
"While the ADAS (Advanced Driver Assistance System) market will take longer to achieve large-scale deployment, I believe solid-state lidar will be used in many high-capacity markets with ADAS applications," said Sinclair Vass, Chief Commercial Officer of Velodyne.
Velodyne was founded in Silicon Valley, USA in 1983. It first started with its audio business, and then expanded its business to lidar and other fields, and is famous for its high-line mechanical lidar. More than half of them have obtained the DMV autonomous driving road test license as its customers.
In 2007, Velodyne released the first commercially available lidar and used it in the map industry in a large number of commercial applications, which is also the starting point of the lidar industry. Velodyne's business involves industrial robots, smart cities and ADAS fields, and partners such as Baidu , Alibaba , Tusen , Pony Ma Zhixing, etc. In October 2020, Velodyne was listed on Nasdaq SPAC, becoming the first lidar stock.
"We don't want to comment specifically on the market price trends in the coming years, but at Velodyne, we recognize that an attractive lidar price will effectively drive the rapid rise of its large number of applications. Our goal is to make lidar available to many end markets through an attractive price roadmap, while we emphasize providing customers with complete solutions, including perception software," said Sinclair Vass.
According to the report of Guojin Securities Research Institute, in the short term, lidar will develop towards hybrid solid state; in the long run, FMCW, OPA, and Flash may all become the dominant routes. There are two major obstacles to the cost and automotive specifications of the laser radar, which can be solved through technological progress and construction of assembly lines.
The perception layer of autonomous driving: "Visualism" and "Lidarism"
"2021 Lidar Application Report on Automotive and Industrial Fields" believes that by 2026, the market size of lidar applied in the automotive and industrial fields is expected to reach US$5.7 billion, with an annual compound growth rate of 21% between 2020 and 2026. It is worth noting that the market share of ADAS (advanced driving assistance system) in 2026 is expected to reach 41%, becoming the largest market segment of lidar, while this data was only 1.5% in 2020.
The three basic elements of autonomous driving are perception-decision-control. A careful perception of the surrounding environment is the basis of all decisions and the safety guarantee for autonomous vehicles.After understanding the position, speed and direction of objects in the surrounding environment, the nature of the road surface, the position of curb , the signal (traffic, road signs), etc., the autonomous driving system needs to start planning and control: first, what other moving objects will do in the next short time, and then plan what you want to do based on the overall plan (such as the planned route to the destination), and finally tell the car what you want to do.
Currently used for sensing mainly includes: lidar, camera, mmWave radar, , ultrasonic radar, etc. Millimeter-wave radar has weaknesses such as being unable to detect pedestrians and stationary objects, while vehicle-mounted camera has weaknesses such as over-reliance on light environments and training samples. In the perception layer of the intelligent driving system, lidar can enhance the redundancy of the perception system, supplement the scenes of missing millimeter-wave radar and cameras, and play a positioning role in conjunction with high-precision maps.
Comparison of different sensors Source: Autobot Reference, China Automobile Center
In the advanced autonomous driving solution, the points of the lidar can also be matched with high-precision map data to position vehicle information in real time. However, there are also problems such as high costs, high impact from bad weather, and short working life, which are expected to be solved through technological progress and large-scale mass production.
As the level of autonomous driving increases, the number of lidar equipment demand increases. Source: Mams Consulting, Guojin Securities Research Institute
has a clear distinction in the perception layer of autonomous driving, "visual school" and "lidar school". This distinction is so well known mainly because Tesla CEO O (Elon Mask) has expressed many times in interviews that "only fools use lidar" and has long insisted on using artificial intelligence and deep learning to build a "pure vision" technical route for neural networks, and even abandoned lidar.
Tesla was the first company to achieve mass production of autonomous vehicles. At that time, a Tesla worth about $70,000 was almost the same as a lidar. Currently, the cost of the Tesla Model 3 ’s autonomous driving camera is only about $65.
Pure visual school believes that relying solely on cameras can complete the surrounding environment perception required for autonomous driving, and prefer pure visual perception solutions such as Tesla, Zekr , and Baidu. The lidar school is dominated by lidar, and uses millimeter wave radar, ultrasonic sensor , and camera multi-sensor fusion to complete the surrounding environment perception. SenseTime AR minibus, Wenyuanzhixing and other uses lidar solutions.
"I think it is OK to achieve autonomous driving if 'vision' can achieve autonomous driving. However, what needs to be considered is what level of safety protection for passengers and road pedestrians after autonomous driving is implemented? How much safety can be improved for road pedestrians? For passengers, can they ride very comfortably in the car and do other things at the same time?" Sinclair Vass put forward this perspective of thinking.
Sinclair Vass believes that if passengers are to have very good comfort and ensure the safety of pedestrians on the road, then it becomes a very important requirement for vehicles to see further and more comprehensively. After all, pure vision has limitations, such as rain and snow, light and different angles. Because of its real-time and comprehensive 360-degree scanning perspective capability, the lidar can see the car further and more comprehensively.
"The autonomous driving technology, which is mainly based on vision, is relatively mature in both hardware and algorithms, and is well understood in terms of advantages and disadvantages and technical bottlenecks. In summary, the technical threshold is relatively low and the price is cheap, but it also faces a relatively large bottleneck. It is difficult to do autonomous driving at high level L3 or above." Sinclair Vass concluded.
Regarding the pure visual route and lidar route of autonomous driving, Li Yikang, director of R&D of SenseTime Intelligent Driving, once told The Paper (www.thepaper.cn), "Whether it is pure visual solution or multi-sensor fusion solution, it is possible to achieve L4 or L5 level autonomous driving. The difference is that the introduction of lidar is actually making the problem simple, because we have introduced a lot of extra information, and this information is very complementary to vision. Some information, such as depth, can be estimated very accurately. If the last two paths can realize L5 level autonomous driving, then I believe that the multi-sensor fusion route may be faster. Of course, perception is only one of the factors that determine whether autonomous driving is implemented."
Editor: Li Yuequn
Proofreading: Zhang Yan
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-