The author recently organized a closed-door discussion with technical experts on 4D imaging radar: Answers to hot topics regarding 4D imaging millimeter-wave radar may all be found here.
Due to the high number of registrations, the first discussion was held last weekend. This face-to-face closed-door discussion lasted for 4 hours, covering four core dimensions of 4D imaging millimeter-wave radar: market, demand, technology, and products, analyzing the industry status, core pain points, and future trends of 4D imaging millimeter-wave radar. Below is the organized discussion minutes; all content comes from frontline development experts and is shared with everyone. (Note: The content only represents the opinions of the experts in this discussion and may not be completely accurate. Please refer to it as industry peers.) There are no pictures in the following text, only textual conclusions (about 8,000 words).
01. Market: Scale, Competition, and Incremental Space
Question 1: What is the market scale and penetration rate of domestic 4D imaging millimeter-wave radar in 2025? What are the future growth trends?
In 2025, the total pre-installed volume of millimeter-wave radar for domestic passenger vehicles reached approximately 36 million units, a year-on-year growth of about 30% from 28 million units in 2024. Among them, 4D-like radars with basic height measurement capabilities accounted for 30%, with an installation volume of about 10.8 million units. Meanwhile, true high-resolution 4D imaging radars had an annual installation volume of about 1.5 million units, accounting for 15% of all 4D-like radars, with a penetration rate of approximately 4.2% in the total millimeter-wave radar market.
According to industry forecasts, over the next 5 years, the proportion of high-end 4D imaging radars among 4D-like radars will increase from the current 15% to 50%-60%, becoming the mainstream solution for automotive millimeter-wave radars. The core drivers are the mandatory requirements of the new national standard for advanced driver-assistance systems (ADAS) and the safety redundancy requirements for L3 and above high-level autonomous driving.
Question 2: How fierce is the price competition in the current automotive millimeter-wave radar market? How much has the cost decreased?
The price war in the automotive millimeter-wave radar market has entered a white-hot stage, and the cost reduction speed far exceeds industry expectations:
Traditional 3D radars: By the end of 2024, the OEM procurement price had fallen below 100 RMB. With the mature solution adopting the TI 2944 chip plus waveguide antennas, there is still room for further price reduction;
Mainstream 8T8R 4D imaging radars: In 2023, OEM quotations were still 1,000-1,500 RMB per unit. In 2025, the mass production supply price for million-level orders has dropped to just over 200 RMB. When OEMs switch from 3D radars to 4D radars, the cost per vehicle increases by only about 140 RMB;
High-end 24T24R 4D imaging radars: The current OEM inquiry range is 600-800 RMB per unit, a drop of over 50% compared to 2023.
A significant part of the price war is because leading manufacturers, in order to seize market share, rush for IPO, or secure OEM design wins, have adopted a "trading price for volume" strategy.
It is understood that a major international manufacturer, to increase its market share, bid for projects at a low price of around 100 RMB; a domestic radar player secured an order of 1.5 million units from an OEM at a price of just over 200 RMB. The industry as a whole has entered a stage of "rising volume but thinning profits."
Question 3: Besides the passenger vehicle pre-installation market, what other deterministic incremental markets are there for 4D imaging millimeter-wave radar?
With fierce price wars and thin profit margins in the main passenger vehicle track, the industry has begun to expand into high-margin segmented scenarios. The core incremental markets include:
Two-wheeled electric motorcycle market: This is currently an important non-automotive incremental scenario, mainly used for BSD (Blind Spot Detection) and RCW (Rear Collision Warning). High-end electric motorcycles have begun to be equipped with forward-facing ACC radars. Some manufacturers can reach an annual cooperative shipment volume of 300,000 units. Although the value per unit is lower than that of automotive applications, they are less sensitive to costs and can charge development fees, contributing stable profits;
Smart home and smart elderly care scenarios: Core applications include indoor human presence detection, fall recognition, and vital sign monitoring such as breathing/heart rate. Compared to cameras, there is no risk of privacy leakage, and compared to 24G radars, the resolution is higher. Currently, 60G/77G radars can accurately distinguish multiple people in meeting rooms, making it a key consumer-level track for the industry;
Industrial inspection and drone scenarios: In the industrial field, it is used for high-precision ranging and material detection; in the drone field, it mainly works with LiDAR to achieve terrain scanning and map building through SAR (Synthetic Aperture Radar) imaging. Although it cannot be used for real-time obstacle avoidance, there are clear demands in agricultural mapping and geographical exploration scenarios;
Low-speed autonomous driving scenarios such as mining trucks and unmanned delivery robots: These scenarios have high requirements for the all-weather working capability of radars and do not require extreme automotive-grade cost control, making them an important survival track for small and medium-sized manufacturers.
Question 4: What is the market competitiveness of domestic millimeter-wave radar MMIC chips? Where is the gap with foreign manufacturers?
The automotive millimeter-wave radar MMIC chip market is still absolutely dominated by foreign manufacturers, and domestic substitution is still in its early stages.
Foreign manufacturer landscape: TI (Texas Instruments), Infineon, and NXP are the three core players, holding a monopoly in the market. One of these manufacturers has even suppressed the survival space of domestic chip manufacturers through significant price cuts. Its reduced prices are now on par with domestic chips, and relying on a mature ecosystem and stable performance, it firmly controls the mid-to-low-end market.
Current status of domestic manufacturers: Caltrate is the only domestic MMIC manufacturer that has achieved large-scale automotive mass production, with large-scale adoption by enterprises such as Chengtai and Huayu; manufacturers such as Haogan Technology, Guibu Micro, Muye Micro, Skyrady, and XJW Micro are also developing 8T8R and higher-channel chips, but none have achieved large-scale automotive mass production.
Core shortcomings: Compared with foreign manufacturers, the products of some domestic players still have gaps in RF performance, automotive-grade certification, and mass production consistency. The technical support capabilities of some domestic manufacturers are also weaker than those of leading foreign enterprises. Some only provide in-depth technical support to top large customers, while small and medium-sized customers can only receive basic materials. There is still a long way to go in ecosystem construction.
02. Demand: Core Demands of OEMs and Algorithm Providers
Question 1: What mandatory demand upgrades has the new national standard for advanced driver-assistance systems brought to 4D imaging millimeter-wave radar?
The new national standard for combined driver-assistance systems is an important policy driver for the rising popularity of 4D imaging radar. The harsh operating condition requirements it proposes cannot be met by traditional 3D 4T4R radars:
Long-distance small target detection: The national standard requires stable recognition of traffic cones, cardboard boxes, and stationary vehicles at a distance of over 160 meters. However, actual test data shows that the stable recognition distance for traffic cones by many leading manufacturers' 4T4R radars is only 20 to 30 meters, and for cardboard boxes, it is only 70 to 80 meters, which completely fails to meet the national standard requirements. In contrast, 4D imaging radars with 6T8R and 8T8R cascaded configurations can achieve stable recognition of over 160 meters, up to a maximum of 200 meters, after algorithm optimization.
Robustness in extreme weather and complex scenarios: The national standard requires stable detection performance to be maintained in rainy days, foggy days, and inside metal-shed tunnels, especially in scenarios such as reversing in tunnels and stationary vehicle recognition. Traditional 4T4R radars can hardly meet the standards in these scenarios, while 4D imaging radars, after joint debugging and optimization with OEMs, can achieve stable detection of over 100 meters. Solutions from some manufacturers can already meet the detection requirements of over 160 meters.
Mandatory safety redundancy requirements: The new national standard's detection requirements for nighttime, low illumination, and small obstacles, as well as the perception redundancy requirements for L3 systems, have driven OEMs to shift from "reducing radar configurations" to "adding radar configurations." BYD proposed a 5-radar solution for its entire vehicle lineup, and OEMs such as SAIC and Geely have followed suit, directly driving the explosion in millimeter-wave radar installation volumes.
Question 2: What are the core functional demands of OEMs and autonomous driving algorithm providers for 4D imaging millimeter-wave radar?
The core demands of OEMs and algorithm providers can be divided into three major dimensions: safety performance, engineering implementation, and commercial cost, with a core focus on two points:
Safety fallback capability for high-level autonomous driving: This is the most core and irreplaceable value of 4D imaging radar. OEMs and algorithm providers explicitly require that under extreme operating conditions where cameras and LiDAR are blinded by dense fog, heavy rain, or strong light, the 4D imaging radar must independently complete the safe pull-over of the vehicle, achieving the minimum level of safety redundancy. To meet this demand, the industry generally believes that radars with 12T16R and above channel counts are needed. The 8T8R solution can barely meet it. The core requirements are stable detection capabilities for curbs, concrete blocks, and stationary obstacles, as well as a low false alarm rate.
High-confidence, low-false-alarm target output: This is the biggest pain point in current algorithm implementation. Some mainstream autonomous driving algorithm providers, due to the high false alarm rate and numerous false points of traditional radars, have set the confidence threshold for radar targets to 0.8, filtering out a large number of valid targets. There have even been cases where, during high-speed rear-end collisions with stationary vehicles, the radar had detected the target but it was filtered out by the algorithm. Therefore, the core demand of algorithm providers is that the targets/point clouds output by the radar must have high confidence and effectively suppress multipath ghosting and false points. At the same time, they require increasing the channel count to 16T16R to improve point cloud density and stability, so that algorithms are willing to truly utilize radar data, rather than just using it as a supplement in the velocity dimension.
Ultimate cost reduction and efficiency improvement: OEMs require 4D radars to control cost increments as much as possible under the premise of meeting performance standards. They explicitly require that the satellite/distributed radar solution must reduce the cost of the MCU (Microcontroller Unit) (about 50 RMB) based on the solution with an MCU, with the target price controlled within 180 RMB. At the same time, they require the standardization and unification of radar interfaces and cameras, sharing deserializers and transmission networks to reduce the supporting costs of the vehicle's electrical and electronic (E/E) architecture.
Question 3: What are the core disagreements in the data application modes of 4D imaging millimeter-wave radar in autonomous driving systems?
The industry's application of 4D radar data is divided into "target-level fusion" or "point cloud-level/raw data early fusion":
Current mainstream: Target-level fusion: Almost all mass-produced vehicle models adopt this mode. Signal processing, target detection, and tracking are completed at the radar end, and only target-level information (position, velocity, confidence, etc.) is output to the domain controller. The required bandwidth is only about 10 Mbps, with no additional requirements for domain controller computing power. Some mainstream algorithm providers currently only use the Doppler velocity information of the radar to correct the velocity detection results of vision, completely without using radar point cloud data. The point cloud is only used for diagnosis and debugging. The advantage of this mode is mature technology and a low implementation threshold, while the disadvantage is that the value of radar data cannot be fully utilized, and it cannot participate in deep fusion.
Future trend: Point cloud-level/raw data early fusion (satellite/distributed radar): This mode removes the MCU at the radar end and transmits the raw ADC data or point cloud data of the radar to the domain controller through a high-bandwidth interface, where the domain controller uniformly completes signal processing and multi-sensor fusion. The advantage is that it can achieve deep early fusion of radar, vision, and LiDAR, with more flexible algorithm iteration and a higher performance ceiling, while also removing the MCU at the radar end to reduce costs. The disadvantage is the high requirement for transmission bandwidth and the need to occupy high computing power of the domain controller (which may require over 10 TOPS). Currently, the computing power of domain controllers is prioritized for vision algorithms, and they are unwilling to allocate too much extra computing power to radar. At present, only a few OEMs such as BYD are promoting mass production, and the industry as a whole is still in the R&D and testing stage.
Question 4: What are the core demands and technical route disagreements of OEMs regarding the interface and communication solutions for 4D imaging millimeter-wave radar?
Regarding radar interface solutions, the core demands of the industry are standardization, high bandwidth, low cost, and high reliability:
SerDes serial interface (current mainstream solution): This solution is the mainstream for automotive sensors. In mass-produced solutions, the forward main radar uses 6G SerDes, and the corner radar uses 3G SerDes, which can meet the transmission needs of radar point clouds and target data. The advantage of this solution is mature technology and a complete supply chain, while the disadvantage is that it only supports point-to-point transmission, does not support network topology and ring network redundancy, and has weak capabilities in time synchronization, diagnosis, and OTA upgrades. Currently, the mainstream is proprietary protocols, with a significant vendor lock-in effect, requiring the entire vehicle to uniformly adopt the solution from the same manufacturer.
Asymmetric Ethernet (IEEE 802.3dm): This standard is expected to be frozen in Q3 2026. Leading manufacturers such as Infineon, ADI (Analog Devices), and TI are all participating in its R&D, planning to release samples in 2027 and achieve mass production in 2028. Its core advantages include: first, the IEEE standard has strict intellectual property reviews, with no patent fees or monopoly risks; second, it natively supports network topology, ring network redundancy, TSN time synchronization, MAC layer security, and whole-vehicle diagnostic OTA, perfectly adapting to zonal E/E architectures; third, it can achieve standardized and unified interfaces with cameras and LiDAR, sharing deserializers and backbone networks to reduce whole-vehicle architecture costs; fourth, the physical layer net rate supports three gears of 2.5G, 5G, and 10G, which can be compared to the available bandwidth of Maxim 12G SerDes, fully meeting the transmission needs of radars and cameras.
Core demands of OEMs: In the short term, prioritize mature SerDes solutions to ensure mass production implementation; in the long term, hope to promote interface standardization, get rid of vendor proprietary standard lock-in, and at the same time require interface solutions to support PoC power supply, remote configuration, and clock recovery to further reduce hardware costs and wiring complexity.
03. Technology: Core Principles, Pain Points, and Breakthrough Directions
Question 1: Compared with traditional 3D radars, LiDAR, and cameras, what are the irreplaceable technical advantages of 4D imaging millimeter-wave radar?
Building on the distance, velocity, and horizontal angle detection capabilities of traditional 3D radars, 4D imaging millimeter-wave radar adds elevation angle detection and high-resolution imaging capabilities. Its irreplaceable advantages over other sensors are mainly concentrated in two points:
All-weather environmental adaptability: Millimeter waves have longer wavelengths and extremely strong penetration capabilities. Under extreme operating conditions such as rain, snow, fog, sandstorms, strong light, and night, the performance degradation is much less than that of LiDAR and cameras. LiDAR and cameras can be blinded in dense fog and heavy rain, while millimeter-wave radars can still maintain stable detection capabilities. This is the core foundation of safety redundancy for high-level autonomous driving and cannot be replaced by other sensors.
Direct, high-precision velocity measurement capability: Based on the Doppler effect, millimeter-wave radars can directly output the radial velocity of targets. The velocity measurement accuracy and response speed are far higher than the indirect velocity measurement methods of cameras through inter-frame difference calculation and traditional LiDAR through point cloud matching calculation. In current mass-produced vehicles, millimeter-wave radar is the only automotive sensor capable of direct velocity measurement. This capability is crucial for the tracking and prediction of dynamic targets, as well as the decision-making and planning of autonomous driving.
In addition, compared with traditional 3D radars, 4D imaging radar solves the pain points of missing height measurement capability, insufficient angular resolution, and weak static target detection capability. It can achieve long-distance detection of small targets such as stationary vehicles, curbs, and traffic cones. At the same time, its cost is much lower than that of LiDAR, and it has higher automotive-grade reliability and mass production maturity, making it the core sensor for achieving high-level ADAS in vehicles priced between 100,000 and 200,000 RMB.
Question 2: What are the core technical pain points of current 4D imaging millimeter-wave radar? What are the mainstream solutions in the industry?
Currently, there are three main pain points for 4D imaging radar:
Core pain point 1: Multipath effect and false point (ghost) suppression
This is the biggest engineering pain point in the industry. After radar electromagnetic waves are reflected by the road surface, guardrails, and surrounding vehicles, mirror false targets are formed, which overlap with real targets. This not only causes false alarms but also drowns out weak real targets.
The mainstream solution for current mass-produced schemes is to solve the multipath problem in typical scenarios on a case-by-case basis through frequency-hopping waveform design, array antenna optimization, multi-frame tracking filtering, and manually rule-based multipath elimination algorithms. This can achieve over 90% multipath false point suppression, but cannot achieve 100% elimination. The frontier exploration directions in the industry include dual-polarization antenna technology and deep learning algorithms to distinguish real targets from mirror targets through multi-dimensional information.
Core pain point 2: The contradictory balance between angular resolution and sidelobe suppression
Angular resolution is a core indicator of 4D imaging radar, which depends on the aperture and channel count of the antenna array. The larger the aperture and the more channels, the higher the angular resolution. However, at the same time, high-resolution sparse arrays will lead to sidelobe elevation. Excessive sidelobes will cause energy leakage of strong targets, creating false points, and will also cause weak targets next to large targets to be drowned out.
The mainstream solution for current mass-produced schemes is to suppress the first sidelobe of the 8T8R solution to 13-14dB through array layout optimization and weighting algorithms, achieving an engineering balance between resolution and sidelobes. At the same time, the industry generally adopts "selective super-resolution algorithms," enabling super-resolution algorithms only in key areas directly ahead to improve resolution, while keeping traditional algorithms in other areas to balance resolution and stability, avoiding detection result jitter caused by super-resolution algorithms.
Core pain point 3: Stable detection of low reflectivity/weak targets
The weak detection capability of millimeter-wave radar for concrete blocks, rough curbs, and low-reflectivity non-metallic targets are a long-standing pain point in the industry. The core reason is that these targets have a high absorption rate for millimeter waves and weak reflected signals, making them easy to be drowned out by noise.
The current mainstream solution is to improve the dynamic range of the radar receiver (the industry generally requires it to reach over 30dB) and optimize the CFAR (Constant False Alarm Rate) detection algorithm to enhance the recognition capability for weak signals. At the same time, the signal-to-noise ratio of weak targets is improved through multi-frame accumulation, balancing detection distance and real-time performance.
Question 3: What is the corresponding relationship between the array channel count of 4D imaging millimeter-wave radar and its performance? Where are the thresholds for entry-level and high-end solutions?
The array channel count is the core hardware foundation determining the performance of 4D imaging radar. Its corresponding relationship with performance and solution thresholds are as follows:
Entry threshold for 4D imaging radar: The industry-recognized entry configuration for 4D imaging radar is 6T8R (6 Transmit, 8 Receive). The core judgment criterion is having effective elevation resolution. Solutions with channel counts lower than this (such as 4T4R) have no effective elevation resolution, can only achieve basic height measurement, and cannot achieve imaging capabilities. They can only be considered "4D-like radars," not true 4D imaging radars.
Corresponding relationship between channel count and core performance:
The more channels, the larger the virtual array aperture. Not only is the angular resolution higher, but the sidelobe suppression capability, anti-interference capability, and point cloud density will also be synchronously improved. Under the same aperture, compared to 16T16R, the core advantage of 24T24R is not a significant improvement in resolution, but better sidelobe suppression, stronger stability and anti-interference capability of detection results, and fewer false points.
Core constraints for channel count improvement: The channel count is not the higher, the better. Its core constraints come from radar size, power consumption, and cost. The installation space for the front bumper of a vehicle is limited, and the radar aperture cannot be infinitely enlarged. The increase in aperture brought by the increase in channel count will directly affect vehicle installation adaptability. At the same time, the more channels, the higher the power consumption. The average power consumption of the 8T8R solution can be controlled within 10W, while the peak power consumption of the 24T24R solution will exceed 12W, putting forward higher requirements for vehicle power supply. In addition, the increase in channel count will directly bring about an increase in chip cost, which is the core bottleneck for the mass production of high-end solutions.
Question 4: What is the current application status and future trend of AI (Artificial Intelligence)/deep learning in 4D imaging millimeter-wave radar?
Currently, the application of AI in 4D imaging radar is still in the R&D exploration stage, with few mass production implementation projects. The industry's application exploration and core bottlenecks are as follows:
Current mass production application status: In mass-produced vehicle models, there is almost no implementation of deep learning algorithms in radar signal processing. Only traditional machine learning methods such as random forests and multi-layer perceptrons are applied in radar target classification and target bounding box regression to optimize target size estimation and classification accuracy. Deep convolutional neural networks are not used.
Core R&D directions in the industry:
Deep learning-based multipath false point suppression: Replacing traditional manually rule-based algorithms, through a data-driven approach, learning the feature differences between real targets and multipath false targets to achieve more universal and efficient ghost elimination. This is the most core R&D direction currently;
End-to-end radar signal processing: Directly outputting target-level information through deep learning networks from raw radar ADC data/range dimension data, replacing the entire traditional process of CFAR detection, DOA (Direction of Arrival) angle measurement, and target tracking, simplifying the algorithm chain, and improving detection performance in extreme scenarios;
Multi-sensor early fusion: Inputting radar point clouds, images, and LiDAR point clouds into a unified deep learning network to achieve end-to-end fusion perception, fully leveraging the velocity measurement and all-weather advantages of radar, and making up for the shortcomings of vision and LiDAR.
Core bottlenecks for implementation:
Difficult data annotation: Radar point clouds are extremely sparse, with only hundreds to thousands of points per frame. Precise manual annotation cannot be done like images. The industry generally uses LiDAR point clouds for benchmark annotation, which is costly and limited in accuracy;
Computing power constraints: The MCU at the radar end has no AI acceleration computing power and cannot run deep learning models; while the computing power at the domain controller end is prioritized for vision algorithms, resulting in obvious resource conflicts.
04. Product: Solution Architecture, Manufacturer Landscape, and Mass Production Implementation
Question 1: What are the mainstream product solutions for currently mass-produced 4D imaging millimeter-wave radar in China?
Currently, the product solutions for 4D imaging radar in China can be divided into three main categories, among which the cascaded solution is the mainstream for mass production, and the distributed solution is the future R&D focus:
6T8R entry-level solution: The industry-recognized entry solution for 4D imaging radar, with costs lower than the 8T8R solution. It is mainly used for forward radars in mid-to-low-end vehicle models and corner radars in high-end vehicle models. It can achieve basic elevation resolution and height measurement capabilities, making up for the shortcomings of traditional 3D radars. Currently, multiple manufacturers have achieved scaled mass production.
8T8R cascaded solution: This solution uses a single 8T8R MMIC chip, which can achieve a horizontal angular resolution of about 2°, an elevation angular resolution of about 3°, and stable detection of small targets over 160 meters, fully meeting the requirements of the new national standard. This solution is the current focus of layout by leading radar manufacturers.
12T16R/24T24R high-end solutions: Belonging to flagship-level 4D imaging radars, they are are mainly used in safety fallback scenarios for L3/L4 high-level autonomous driving. Among them, HELLA's 24T24R solution has achieved a design win at Volkswagen in Germany. Domestic manufacturers are all in the testing and validation stage. OEMs have begun to conduct real-vehicle performance acceptance jointly with manufacturers, and mass production implementation is expected in 2026-2027.
Satellite/distributed radar solution: The core feature is removing the MCU at the radar end, keeping only the MMIC and RF front-end, and transmitting raw data to the domain controller for centralized processing. BYD is the most active OEM in promoting this solution in China. Several leading radar manufacturers have participated in the project R&D. It is currently in the initial stage of mass production. The core cost reduction goal is to control the price of a single radar within 180 RMB, which is the core product R&D direction for the industry in the next 2-3 years.
Question 2: What is the current competitive landscape of 4D imaging millimeter-wave radar manufacturers in China?
The domestic 4D imaging millimeter-wave radar market has formed a clear echelon pattern, with significant differences in core advantages and survival strategies among manufacturers in different echelons:
Local leading manufacturers with large-scale mass production implementation. Representative enterprises: SenSik, Sain Lingdong, etc. They have achieved large-scale pre-installation mass production of 4D imaging radar, and their algorithms have been polished by massive real-vehicle data, making them the mainstream choice for current OEMs. The core strategy of such manufacturers is "trading volume for price," thinning costs through large-scale orders, while laying out high-end solutions and satellite radars to lock in future high-end projects of OEMs.
International Tier 1 manufacturers. Representative enterprises: Continental, Bosch, HELLA, Aptiv. The core advantage is deep technical accumulation and mature high-end solutions. HELLA's 24T24R solution has achieved design wins overseas, and the 3D radars of Continental and Bosch hold a monopoly in the domestic market. However, the shortcoming of such manufacturers is high prices and unwillingness to participate in low-price competition in the domestic market. In domestic design win projects for 4D imaging radar, they have been surpassed by local manufacturers, maintaining advantages only in joint-venture brand vehicle models.
Local manufacturers without scaled implementation. Such enterprises have complete technical reserves and R&D capabilities but are in a passive position in market competition. Some enterprises, insisting on "not doing unprofitable business," are unwilling to bid at low prices and have no mass production design wins in the passenger vehicle 4D radar market. Such manufacturers are also laying out non-automotive tracks to balance profitability and R&D investment, facing greater survival pressure in the future.
Start-up manufacturers and segmented track players. Such manufacturers avoid the price war in the main passenger vehicle track, focusing on segmented scenarios such as two-wheelers, industry, smart homes, and smart elderly care. Relying on customized solutions and flexible services, they achieve stable shipments in segmented markets and are an important supplement to the industry. The core survival logic of such manufacturers is to find high-margin incremental markets outside the main track to avoid direct competition with leading manufacturers.
Question 3: What is the future product evolution direction of 4D imaging millimeter-wave radar?
Based on the consensus of frontline industry discussions, over the next 3-5 years, the product evolution of 4D imaging millimeter-wave radar will revolve around three core directions:
Ultimate cost optimization, with satellite/distributed architectures gradually achieving mass production: With the popularization of central computing E/E architectures, satellite/distributed radars will move from R&D to mass production. The hardware cost of radars after removing the MCU will further drop, and the price of entry-level 4D radars is expected to fall below 150 RMB, becoming standard equipment for 100,000 RMB-level vehicle models.
Continuous performance upgrades, with steady improvement in channel count: The 8T8R solution will maintain its mainstream mass production position for a long time in the future, while 12T16R and 16T16R solutions will become standard for L3-level autonomous driving. Solutions with 24T24R and higher channel counts will be implemented in L4 autonomous driving scenarios. At the same time, the angular resolution and point cloud density of radars will continue to improve, gradually approaching the imaging effects of low-line LiDAR, achieving blind-spot filling for LiDAR in more scenarios, and even replacing LiDAR in some mid-to-low-end vehicle models.
Deep integration of algorithms and architectures, moving from "sensors" to "perception systems": The algorithms of 4D radar will move from the radar end to the domain controller end, deeply coupling with the fusion perception algorithms of autonomous driving, upgrading from simple hardware sensors to core links in the autonomous driving perception system. At the same time, deep learning algorithms will gradually be implemented in specific scenarios to solve core pain points such as multipath suppression and weak target detection. The false alarm rate and detection stability of radars will achieve qualitative improvements, truly becoming an indispensable safety redundancy core for high-level autonomous driving.
05. Continued Discussion
Related topics will continue to be discussed recently. If you are interested in 4D imaging millimeter-wave radar technology, you are welcome to participate and communicate together. Registration address: https://dcn7get8fskg.feishu.cn/wiki/LaNVwbdFeizESMkI04UcMaBcnyb
Personal opinions, not necessarily accurate, welcome to discuss. I am Xueling, researching the technology, products, and applications of AI (Artificial Intelligence), welcome to communicate.