2021 was dubbed the first year of intelligent vehicles, with autonomous driving technology receiving unprecedented investment from all parties.
Whether it's traditional automakers or emerging car manufacturers, whether it's mass-produced cars or concept cars, when automakers claim that their autonomous driving technology has reached L2, L3, or even L4 levels, we can vaguely sense that the competition between software and hardware in car manufacturing has entered a fierce stage.
Among these, one hardcore component has frequently appeared in the configuration lists of major automakers regarding autonomous driving. The Jihu Alfa S, XPeng P5, NIO ET7, SAIC RS33, BMW iX, and Zhiji L7 have all announced that they will be equipped with this application. The number and amount of financing in this sector have repeatedly reached new highs, and it has become the norm for related concept stocks to soar tenfold. This component is LiDAR.
With tech giants pouring capital, traffic, and technology into the field, the commercialization and scaling of autonomous driving are accelerating, creating a blue ocean market for LiDAR. Statistics show that at least 20 automakers and autonomous driving companies have announced that they will incorporate LiDAR as part of their perception suites, aiming to mass-produce Level 3 and above autonomous driving technology. It can be said that LiDAR has now become a standard feature for intelligent vehicles achieving high-level autonomous driving. Of course, besides Tesla. Amidst this wave, a quiet battle is brewing within the industry regarding LiDAR.
PART 01
Who is the eye of intelligent driving?
In "Fast & Furious 8," there is a memorable scene involving autonomous driving: the main villain, in order to seize a nuclear weapon launch device, uses high technology to hack into the intelligent driving center system of cars, controlling thousands of driverless cars to form a massive "driverless car" army. The car army maneuvers through the streets in an orderly fashion to intercept the Russian Minister of Defense's car. The car army moves rapidly through the crowded streets, creating a stunning scene.
But back to reality, is this level of autonomous driving even possible for us at present?
Before answering this question, we first need to understand the internationally recognized classification standards for driving automation.
According to the latest version of the autonomous driving standard jointly released by the Society of Automotive Engineers (SAE) and the International Organization for Standardization (ISO), the capabilities of driver assistance and autonomous driving systems are divided into six levels, ranging from L0 (no driver automation) to L5 (fully automated). This new framework clarifies the difference between Level 2 and Level 3, naming Level 0-2 systems "driver assistance systems," while Levels 3 to 5 are considered "autonomous driving systems."
▲SAE Driving Automation Classification Standard
The new regulations stipulate that Level 3 and Level 4 autonomous driving systems can drive vehicles under limited conditions, but will not operate autonomous driving unless all required conditions are met; Level 5 autonomous driving systems can drive vehicles under all conditions.
According to this standard, the "driverless car" army in *Fast & Furious 8* should be at Level 5. However, in reality, aside from promotional hype and differences in standards, the actual autonomous driving capabilities of most leading automakers are generally between Level 2 and Level 3, still a huge gap from Level 5. Even Tesla, previously considered a leader in autonomous driving, only has an actual autonomous driving capability around Level 2, although the stubborn Musk refuses to admit it.
How to achieve higher levels of autonomous driving capabilities? This mainly depends on three key components of the intelligent driving module: environmental perception, computational decision-making, and control execution, which correspond to the human "eyes-brain-nerves," respectively. Leaving the latter two aside for now, let's focus solely on the mainstream visual recognition solutions on the market. The industry's disagreement lies in which of the following: cameras, millimeter-wave radar, ultrasonic radar, and lidar—is the true "eye of intelligent driving"?
PART 02 LiDAR in the Mid-Range Battlefield: Cameras, millimeter-wave radar, ultrasonic radar, and LiDAR are currently the four most commonly used sensor solutions in the field of autonomous driving. They differ in detection parameters such as detection range, resolution, and angular resolution, and each has its own advantages and disadvantages in object detection, recognition and classification, 3D modeling, and resistance to adverse weather conditions. It is worth noting that these four sensor solutions, in automotive-grade applications, are not isolated or independent, but rather complement each other, leveraging each other's strengths and weaknesses.
▲ Comparison of four mainstream autonomous driving detection sensors
Currently, mainstream autonomous driving manufacturers generally follow two different paths in their selection of autonomous sensors: one is pure vision computing, which is dominated by cameras and combined with low-cost components such as millimeter-wave radar, with typical representatives being Tesla and Mobileye; the other is dominated by LiDAR, combined with cameras, millimeter-wave radar and other components, with typical representatives being Google Waymo, Baidu Apollo, and Pony.AI in China.
To make a simple analogy, the visual solution imitates the "human eye," which mainly relies on seeing, and is equivalent to a two-dimensional camera; the radar solution imitates a bat, which is a three-dimensional "scan," and is equivalent to a three-dimensional scanner.
Currently, the vision-based approach, primarily using cameras and millimeter-wave radar, has reached a fairly mature level of L2 autonomous driving technology. Its algorithms have also made significant progress through continuous optimization, with Tesla being a prime example. However, cameras capture two-dimensional images, which are not only more difficult to identify and process than three-dimensional images, but also require more powerful algorithms, extensive data training, and longer-term R&D investment. Furthermore, they suffer from limitations in accuracy, stability, and field of view, and are currently unable to meet the requirements of L3 and higher levels of autonomous driving.
▲Comparison of sensor performance
To achieve higher levels of autonomous driving capabilities, the mainstream view in the industry is that LiDAR is the core hardware for developing fully autonomous driving technology.
Compared to autonomous driving solutions that rely heavily on computing and lightly on perception, the most obvious characteristic of autonomous driving solutions that rely heavily on LiDAR is their emphasis on perception and light on computing. Significantly improving the perception of road conditions and the environment by stacking LiDAR sensors on the vehicle is a necessary condition for accelerating the commercialization of Level 4 and above autonomous driving capabilities. Brad Templeton, a Google autonomous driving consultant, once pointed out incisively, "Achieving 99% accuracy is not enough for vehicle driving; we need 99.99999% accuracy. LiDAR provides the strongest guarantee to several decimal places." Currently, there is little debate in the academic field about whether LiDAR is advanced. The principle in academia is that a solution that provides more and more accurate data is a better solution.
Many autonomous driving technology companies share a consensus: the first half of the LiDAR deployment in vehicles has completed the verification process from 0 to 1, but the second half, from 1 to N, still requires concrete commercial practice. In other words, LiDAR has achieved small-scale deployment at the experimental level, proving the technology's feasibility; the subsequent large-scale mass production deployment is the only standard to verify its commercial success.
▲ LiDAR equipment status for some models
Since the beginning of 2021, major emerging electric vehicle manufacturers have accelerated the commercialization of LiDAR solutions. XPeng P5, NIO ET7, Zhiji L7, WEY Mocha, Jihu Alfa S, and BMW iX have all announced that they will equip their new vehicles with LiDAR (as shown in the image above). Among traditional automakers, Audi, Nissan, and Toyota have also announced plans to implement LiDAR in their vehicles, but specific timings have not yet been announced.
With the large-scale commercialization of autonomous driving, LiDAR will also usher in its opportunity to shine after a long process from 0 to 1. However, in this brief intermission from 1 to N, the selection of core technology routes remains a difficult problem facing industrial implementation.
PART 03 The "Triple Threshold" of Technical Approaches Because the components required for LiDAR to function are numerous and their principles differ to varying degrees, there are multiple mainstream classification methods and significant differences in technical paths. LiDAR companies employ different technical approaches to compete. Based on structure, LiDAR can currently be broadly divided into three types: mechanical, semi-solid-state, and solid-state, which can be considered as corresponding to the test version, market version, and high-end version of the technical approach, respectively.
▲ LiDAR structure classification diagram
Mechanical LiDAR is currently the most mature type, mainly collecting information by mechanically rotating 360° to scan. The faster the rotation speed, the more information is collected. However, due to the large number of mechanical parts, it is bulky, complex to assemble, has a long production cycle, and is prone to damage in real road conditions, resulting in high installation costs. Therefore, it is difficult to meet the requirements of automotive-grade LiDAR.
Semi-solid-state LiDAR is currently the most promising solution for rapid deployment. It is mainly divided into two types: rotating mirror and MEMS. By integrating mechanical components onto a single chip and controlling the rotation through circuitry, it not only simplifies the mechanical components and reduces the size, but also greatly reduces costs and improves mass production capabilities. Although it sacrifices some performance in detection effect, making it inferior to mechanical LiDAR, it is still the mainstream choice for autonomous driving.
Solid-state LiDAR represents the ultimate vision for LiDAR technology, primarily comprising phased array and Flash technologies. Neither requires mechanical components, but both are currently immature. The former's technology operates on a principle similar to phased array radar in fighter jets, resulting in high mass production costs. The latter utilizes a fast-flash principle, capable of completing a 3D rendering of the entire scene in a single operation. It's fair to say that neither of these technologies will be readily adopted for large-scale automotive-grade applications in the near future.
Overall, the high cost, low reliability, and lack of mass production capability of mechanical LiDAR directly restrict its mass production and large-scale application; semi-solid-state LiDAR is easier to pass automotive-grade standards, and the rotating mirror solution is about to be mass-produced, making it the main choice for automakers to install in vehicles; although solid-state LiDAR can achieve very low cost, the technology and industrial chain are still immature, and there is still a certain distance to go before it can be mass-produced.
In recent years, LiDAR technology has developed rapidly, becoming a highly sought-after investment. Velodyne, Luminar, Hesai, Innoviz, and Aeva have all raised billions of RMB in total funding. Since 2019, more than 13 domestic LiDAR companies have secured new rounds of financing. However, almost all of these companies are still operating at a loss. To achieve profitability, mass production and vehicle deployment are inevitable, but cost remains a major obstacle for automakers.
▲The Impossible Triangle of Industrial Manufacturing
In industrial product manufacturing, there is an impossible triangle: performance, price, and size; at most, only two of these three can be satisfied. The same applies to lidar. From mechanical and semi-solid-state to solid-state, lidar technology has continuously evolved. This evolution is not linear, but rather involves constantly balancing and making trade-offs within the impossible triangle to find the optimal center point, thereby achieving feasible commercialization.
Li Zhenyu, senior vice president of Baidu Group and general manager of the Intelligent Driving Business Group, said in an interview: "High-beam LiDAR still faces the issues of automotive standards and cost. It is difficult to say whether the cars using LiDAR on the market are demos or real mass-produced cars. However, if LiDAR is to be truly implemented, cost reduction is an inevitable trend."
Early mechanical LiDAR, such as Velodyne's, cost around $80,000 for its 64-line mechanical LiDAR, $20,000 for its 32-line LiDAR, and even the cheapest 16-line LiDAR cost $4,000 in 2018. SCALA, the first automotive-grade LiDAR, also reached a price of around $20,000 in its first generation. Considering that the current Tesla Model 3 only costs $35,000, a single LiDAR device accounted for a large portion of the vehicle's cost, even exceeding the price of the car itself. No wonder Musk famously declared that "any company relying on LiDAR will fail." After all, the Model 3's autopilot camera only costs $65. However, with technological advancements, the cost and size of LiDAR have decreased significantly. Since 2020, semi-solid-state and solid-state LiDAR have gradually replaced mechanical LiDAR, becoming smaller and cheaper, dropping from tens of thousands of dollars to around $1,000, gradually becoming affordable for automakers. This has led to the current situation where LiDAR is increasingly being used in vehicles.
▲Prices of some lidar products
According to the latest McKinsey Automotive Consumer Insights 2021 report, 90% of respondents believe that driver assistance systems are meaningful, and 10%-35% of consumers are willing to pay RMB 2,200-4,100 for Level 2 driver assistance systems. Level 2.5/Level 3 systems are even more valuable, with 15%-30% of respondents having a price range of RMB 3,800-4,900.
As shown in the above figure, current LiDAR solutions still fall short of consumers' expected price range. With autonomous driving becoming an increasingly competitive area for automakers, the technological evolution of LiDAR will accelerate further, and costs will decrease. It is expected that semi-solid-state and solid-state LiDAR will see mass-produced automotive-grade applications within the next 3-5 years. PART 04 Overseas First-Mover Advantage, Domestic Rise As an emerging field, LiDAR is a blue ocean market globally. However, in terms of technology development and industrial application, overseas manufacturers have a first-mover advantage over domestic manufacturers in both upstream and downstream industries. Domestic manufacturers have been catching up rapidly in recent years, achieving many breakthroughs, and China's strength is gradually rising. According to the goals set in the "Intelligent Connected Vehicle Technology Roadmap 2.0," from 2020 to 2025, the sales volume of L2-L3 level intelligent connected vehicles in China will account for more than 50% of the total annual vehicle sales, L4 level intelligent connected vehicles will begin to enter the market, the C-V2X terminal installation rate in new vehicles will reach 50%, and L4 level vehicle commercial applications will be carried out in specific scenarios and limited areas. By 2026-2030, the sales share of L2-L3 level intelligent connected vehicles will exceed 70%.
In recent years, the Chinese government has successively introduced a series of policies to promote the development of autonomous driving, which has further boosted the development of my country's LiDAR industry. By the end of 2019, 25 cities across the country had issued autonomous driving policies; in February 2020, the National Development and Reform Commission (NDRC) led the release of the "Intelligent Vehicle Innovation Development Strategy"; in 2020, the NDRC for the first time clearly defined the seven major sectors of "new infrastructure," and LiDAR, as a terminal sensing device, has played an increasingly prominent role in intelligent transportation and smart cities, including autonomous driving and vehicle-road cooperation.
According to statistics, there are currently more than 50 domestic players in the LiDAR market. Among them, three well-known companies entered the market earlier, including RoboSense, Hesai Technology, and LeiShen. In addition, there are also startups focusing on cutting-edge technologies, such as Moore Core Optoelectronics, Beike, Feixin Electronics, and Beike Tianhui.
RoboSense and Hesai Technology are representative domestic manufacturers that have chosen the same development path as Velodyne. In the field of mechanical LiDAR, domestically produced LiDARs with the same line count are 1/3 to 1/2 cheaper than Velodyne's. For top-of-the-line 64-line and 128-line products, Velodyne's 64-line LiDAR costs 500,000 to 600,000 yuan, while Hesai's 64-line product costs only over 200,000 yuan. In the 16-line LiDAR market, Velodyne's products cost tens of thousands of yuan, while RoboSense's similar products only cost 20,000 to 30,000 yuan.
Moore Optics possesses world-leading Optical Phased Array (OPA) technology and has completed a laboratory demo, placing it at the forefront of chip-level solid-state LiDAR R&D. Compared to traditional solutions, chip-based solutions use fewer light sources and cheaper detectors, saving significant costs in the packaging stage. The company expects to achieve mass production of highly reliable, low-cost silicon photonics LiDAR by 2025. Feixin Electronics' innovative 3D Flash automotive radar solution can eliminate crosstalk and background light, resist interference, and achieve high-precision long-range detection. The company's full product line will be released successively in the second half of this year, and it has already cooperated with leading customers in automotive and consumer applications, with expected shipments of nearly one million units (sets) in 2021. Beike Tianhui's new generation CFans-32 and CFans-128 solid-state LiDAR products will comprehensively upgrade the sensing technology of intelligent automotive LiDAR, providing intelligent vehicle navigation with comprehensive high-precision 3D target detection and classification capabilities. In the future, with the further popularization of autonomous driving technology, the LiDAR market size will further expand. According to consulting firm Yole, LiDAR applications are one of the fastest-growing sectors in the automotive industry. In terms of shipments, Yole predicts that global LiDAR shipments will reach approximately 4.7 million units in 2025 and 23.9 million units in 2030; in terms of sales revenue, global LiDAR sales are projected to reach approximately $6.19 billion in 2025 and $13.932 billion in 2030. Overall, after years of ups and downs, LiDAR has seen its first large-scale industrialization this year, riding the wave of "autonomous driving," but at the same time, the industry is undergoing rapid consolidation. Among the remaining players are seasoned veterans, as well as aggressive tech giants and emerging companies with cutting-edge technologies.
The debate over LiDAR technology routes is still ongoing and highly contentious. Whether traditional mechanical, semi-solid-state, or all-solid-state, each technology still faces numerous unresolved issues. Looking at the broader context of autonomous driving, while the anticipated Level 4 autonomous driving is unlikely to see widespread adoption in the short term, the gradual mass production of Level 3 autonomous vehicles is creating a surge in demand for LiDAR. It is foreseeable that as autonomous driving accelerates, the integration of LiDAR into vehicles will become an inevitable trend, and the evolution of technological routes will inevitably accelerate a major industry reshuffle. Regardless of scale, funding, or technological leadership, when the tide recedes, those who best adapt to the market will survive.
