2025년 초, BYD의 Eye of God Intelligent Driving과 Changan Automobile의 Tianshu Intelligent Driving이 대중을 위한 지능형 운전의 물결을 일으켜, 지능형 운전의 민주화가 점점 분명해졌습니다. LiDAR 기술은 현재 10만 위안-15만 위안의 모델(Galaxy E8, bZ3X, Leapmotor B10 등), 심지어 10만 위안의 모델(Changan Automobile의 모델이 탑재 예정)까지 확대하고 있습니다.
또한 NIO ET9(3개의 LiDAR), MAEXTRO S800(4개의 LiDAR), New AITO M9(4개의 LiDAR) 등의 하이엔드 모델은 안전 중복성을 높여줍니다. Zeekr Qianli Haohan H9는 5개의 LiDAR을 탑재하고 올해 4분기에 출시 예정인 GAC 그룹의 L3 자율주행 모델 G1000은 4개의 LiDAR을 탑재합니다. L3/L4 자율주행으로 진행하려면 고성능 LiDAR로의 업그레이드와 그 수 증가가 필수적입니다.
차종의 부침뿐만 아니라 성능 향상과 비용 절감의 부침도 있어 휴머노이드 로봇, 로봇 개, 저고도 경제, 물류, 항만, 농업 등, 보다 많은 장면에서의 LiDAR의 응용도 촉진되고 있습니다. LiDAR은 자동차 및 비자동차 분야에서 모두 폭발적인 보급을 보여줍니다.
ResearchInChina의 통계에 따르면 2024년 LiDAR 탑재수는 전년 대비 245.4% 증가한 152만 9,000대로 급증했고 보급률은 6.0%로 급상승했습니다. LiDAR을 탑재한 모델은 시장에서 점점 인기가 높아지고 있습니다.
시장 집중도라는 점에서 자동차용 LiDAR의 상위 4개사는 RoboSense, Hesai Technology, Huawei, Seyond입니다. 2024년, 이 4개 회사의 총 시장 점유율은 99%를 넘어 자동차용 LiDAR 시장을 독점했습니다. 기타 승용차용 LiDAR 공급업체에는 Luminar, Valeo, DJI Livox, Tanway Technology 등이 있으며 이러한 공급업체도 대량 생산을 실현하고 있습니다.
LiDAR의 칩화는 합리화된 폼 팩터(루프 탑재형), 컴팩트한 통합(앞 유리 뒤, 범퍼 뒷면, 헤드라이트와의 융합), 보다 세밀한 환경 인식을 가능하게 함으로써 안전 중복성 강화 등 엔지니어링 수요에 부응하는 것입니다. 발광, 스캔 및 수신의 각 구성 요소를 소형화하고 통합함으로써 비용을 더욱 절감할 수 있습니다. 동시에 중앙 집중식 컴퓨팅과 같은 디지털 아키텍처 설계는 온보드 응답 시간을 가속화하고 AEB 최적화와 같은 안전 기능을 향상시킵니다.
칩화의 명확한 동향에도 불구하고, 기술적인 병목 현상은 여전히 남아 있습니다. SPAD 칩 기술은 여전히 Sony 및 ON Semiconductor와 같은 국제 기업이 독점하고 있으며, 실리콘 포토닉 OPA의 스캔 정밀도는 더욱 개선이 필요합니다. 국내 제조업체는 완전한 공급망 자립을 달성하기 위해 재료(InGaAs 검출기 등) 및 공정(3D 적층 등)의 획기적인 가속을 가속시켜야 합니다.
예를 들어, Aeva의 Atlas Ultra LiDAR의 발전은 Aeva CoreVision(TM) LiDAR 칩 모듈과 Aeva X1(TM) SoC 프로세서를 포함한 맞춤형 실리콘에 의존합니다. 4세대 CoreVision 모듈은 이미터, 검출기, 광학 인터페이스 칩과 같은 중요한 LiDAR 부품을 단일 자동차 등급 설계로 통합합니다. 독자적인 실리콘 포토닉스를 활용하여 복잡한 광섬유를 대체하여 품질이 뛰어난 확장 가능하고 비용 효율적인 대량 생산을 실현합니다.
이 보고서는 중국의 자동차용 LiDAR 산업에 대한 조사 분석을 통해 LiDAR 개요, 응용 시나리오, 국내외 공급업체, 개발 동향 등의 정보를 제공합니다.
In early 2025, BYD's "Eye of God" Intelligent Driving and Changan Automobile's Tianshu Intelligent Driving sparked a wave of mass intelligent driving, making the democratization of intelligent driving increasingly evident. LiDAR technology has now been extended to models priced between 100,000 and 150,000 yuan (such as the Galaxy E8, bZ3X, and Leapmotor B10), and even to a 100,000-yuan model (a Changan model will be equipped with it).
Additionally, high-end models such as the NIO ET9 (3* LiDAR), MAEXTRO S800 (4* LiDAR), and New AITO M9 (4* LiDAR) are enhancing safety redundancy. The Zeekr Qianli Haohan H9 will be equipped with 5* LiDAR, while the GAC Group's L3 autonomous driving model G1000, set to launch in Q4 this year, will feature 4* LiDAR. Upgrading to high-performance LiDAR or increasing their number has become essential for advancing to L3/L4 autonomy.
Beyond the ups and downs in vehicle models, there is also an ups and downs in performance improvement and cost reduction, which also promotes the application of LiDAR in more scenarios, such as humanoid robots, robot dogs, low-altitude economy, logistics, ports, agriculture, etc. LiDAR is experiencing an explosion in both automotive and non-automotive fields.
According to statistics from ResearchInChina, the installed capacity of LiDAR surged to 1.529 million units in 2024, a year-on-year increase of 245.4%; the penetration rate rapidly jumped to 6.0% in 2024. Models equipped with LiDAR are becoming more and more popular in the market.
In terms of market concentration, the top four automotive LiDAR companies include RoboSense, Hesai Technology, Huawei, and Seyond. In 2024, the combined market share of these four companies exceeded 99%, dominating the automotive LiDAR market. Other passenger car LiDAR suppliers include Luminar, Valeo, DJI Livox, Tanway Technology, etc., which are also achieving mass production.
LiDAR chipification addresses engineering demands for streamlined form factor (roof-installed), compact integration (behind windshield, bumper, or fused with headlights), and enhanced safety redundancy by enabling finer environmental perception. By miniaturizing and integrating emission, scanning, and reception components, it further reduces costs. Concurrently, digital architecture designs, such as centralized computing, enable faster onboard response times, improving safety features like AEB optimization.
Despite the clear trend toward chipification, technical bottlenecks persist: SPAD chip technology remains dominated by international players like Sony and ON Semiconductor, while silicon photonic OPA scanning accuracy requires further refinement. Domestic manufacturers must accelerate breakthroughs in materials (e.g., InGaAs detectors) and processes (e.g., 3D stacking) to achieve full supply-chain autonomy.
For example, Aeva's Atlas Ultra LiDAR advances rely on custom silicon, including the Aeva CoreVision(TM) LiDAR chip module and Aeva X1(TM) SoC processor. The fourth-gen CoreVision module integrates all critical LiDAR components - emitters, detectors, and optical interface chips - into a single automotive-grade design. Leveraging proprietary silicon photonics, it replaces complex fiber optics, ensuring quality and scalable, cost-effective mass production.
Additionally, Aeva X1, a FMCW LiDAR SoC processor, seamlessly integrates data acquisition, point-cloud processing, scanning systems, and application software into a single mixed-signal chip.
RoboSense restructured LiDAR architecture through chipification, consolidating discrete components into chips to slash assembly costs. Its MX product, for instance, replaces FPGAs with ASICs, reducing costs to under USD200 and enabling adoption in RMB150,000-RMB200,000 vehicles.
The MX also features RoboSense's self-developed SoC, the M-Core, with powerful processing capabilities and multi-threshold TDC (Time-to-Digital Converter), boosting weak-echo detection by 4 times and range resolution by 32 times. RoboSense has achieved chipified scanning, integrated data processing, and iterative transceiver upgrades.
Hesai's AT512 LiDAR employs chipified control to achieve 400m detection range while improving optical efficiency via integrated VCSEL and single-photon detectors.
In January 2025, Hesai launched the world's first 1,440-channel ultra-long-range LiDAR, powered by its Gen4 chip. It leverages advanced high-efficiency sensing and ultra-parallel processing to deliver unprecedented perception, producing image-grade point clouds that capture road imperfections, pedestrians, and vehicle details with precision.
Key features of Hesai's Gen4 chip include: 1. 3D stacking technology enabling single-board integration of 512 channels. 2. A 256-core Intelligent Point-cloud Engine (IPE) and 8-core APU, achieving 24.6 billion samples per second. 3. 130% higher detector sensitivity and 85% lower per-point power consumption. 4. Support for all-solid-state e-scanning, photon anti-interference, and smart optical zoom.
Digitalization is also a key focus in the LiDAR industry. Digital LiDAR employs digital methods to detect and process photon information, eliminating the "analog-to-digital" conversion process. This preserves more detection data, enhances resolution, accuracy, integration, and perception fusion capabilities, while delivering additional system-level benefits.
Digital LiDAR utilizes Single-Photon Avalanche Diode (SPAD) devices, which detect laser signals at the single-photon level. The output digital signals can proceed directly to processing without requiring intermediate transmission components. Meanwhile, signal processing, storage, and even laser control can be integrated into chips via algorithms, improving computational efficiency while reducing reliance on physical hardware.
Current SPAD chip players include Sony, as well as domestic entrants like Sophoton, FortSense, and Adaps Photonics. Companies adopting SPAD-based digital architectures include Ouster, ZVISION, and RoboSense. For example, the ZVISION EZ6, which uses SPAD chips, achieves a 20%-30% cost reduction compared to previous generations, making it suitable for forward long-range applications (passenger cars/intelligent transportation).
The EM4, the first product under RoboSense's new digital EM platform, integrates a SPAD-SoC chip and a 940nm VCSEL chip. As the world's first 1080-channel LiDAR, it can precisely identify distant small objects like tires, traffic cones, and cartons, raising the safety ceiling for autonomous driving systems. It can improve system response time by up to 70%, enabling more confident decision - supported by direct integration with automotive Ethernet systems in smart vehicles. RoboSense's digital LiDAR will accelerate adoption across automotive, robotics, and drone markets.
In terms of algorithm and architecture innovation, take VanJee Technology's 192-channel LiDAR WLR-760 as an example. It adopts a VCSEL+SPAD design, combined with VanJee's self-developed FOC vector control algorithm for rotating mirrors and multi-channel VCSEL drivers. This not only significantly improves product performance but also simplifies the internal structure. Compared to traditional solutions, the number of component types is reduced by over 60%, the quantity of components by over 80%, and production steps by 30%.
In information processing, there is a trend toward shifting computing power upward. For instance, ZVISION's SPAD product architecture retains only the optoelectronic front end, transmitting raw signals directly to the domain controller. The EZ-Key algorithm suite is deployed on the domain controller side, moving LiDAR computing tasks to the domain controller. This approach modularizes the LiDAR's optoelectronic front end, minimizes its power consumption, and standardizes LiDAR data. It also enables the use of massive amounts of raw corner-case data to iteratively upgrade point cloud algorithms.
The EZ-Key suite can be deployed either on the LiDAR unit itself or flexibly integrated into customer's domain controller. Its functions include dirt detection, rain/fog/dust/exhaust detection, line-drawing algorithms, ghosting removal algorithms, and bloated-point suppression algorithms, effectively addressing the impact of false point clouds on data quality in various scenarios.
As LiDAR point cloud quality approaches the pixel-level clarity of cameras, and with LiDAR's zoom capability mirroring that of camera lenses, parameters can be dynamically adjusted based on driving scenarios and needs. This enhances recognition in the central field of view, with finer resolution for clearer perception. For example, the focal length can be extended for highway driving to detect distant obstacles earlier, or narrowed for urban congestion to better perceive nearby vehicles and pedestrians. Zoom-capable LiDARs have already been deployed in mass-produced models like the Hyptec GT.
For cost reduction, companies can improve system integration through chip-based and digital architecture designs. By enhancing production processes and introducing highly automated equipment, they can cut labor calibration costs-which account for about 20% of LiDAR costs. Additionally, higher integration reduces the number of key suppliers, improving supply chain stability and enabling faster large-scale automation, further lowering manufacturing costs.
Economies of scale drive marginal cost reductions. Hesai plans to deliver 1.2 to 1.5 million LiDAR units in 2025, with over 80% allocated to ADAS applications. RoboSense aims to penetrate the mid- to low-price vehicle market in 2025 with its MX series (priced below $200) and accelerate expansion into emerging sectors like robotics and industrial applications.
In non-automotive applications, LiDAR is being widely adopted in industrial control, robotics, drones, measurement & ranging, ports, logistics, agriculture, and other sectors. In December 2024, Hesai delivered over 20,000 LiDAR units for the robotics market in a single month.
Hesai Technology stated that its LiDAR shipments in 2025 are projected to reach 1.2 to 1.5 million units, with approximately 200,000 units designated for robotics applications-covering mobile robots, delivery robots, cleaning robots, and more. Its new production line is set to commence operations in Q3 2025, with annual capacity expected to reach 2 million units by year-end. Hesai's XT series currently provides 3D perception technology for Unitree's robots and is deployed in scenarios such as BMW's Automated Factory Driving (AFD) system.
Meanwhile, RoboSense officially announced its robotics platform company strategy in early 2025, positioning itself as a "robotics technology platform company" to supply incremental components and solutions for the AI robotics industry. Products like the E1R and Airy LiDARs for robots, along with new robotics vision offerings such as the Active Camera and the dexterous hand Papert 2.0, are rapidly being implemented in AI robotics applications.
Seyond is also actively expanding in the robotics market, with its products already deployed across major applications including robotic dogs, logistics robots, industrial robots, and agricultural robots. The company continues to see growing shipments in this sector.
Finally, let's examine how other LiDAR companies are advancing product applications in non-automotive fields.