In the realm of autonomous driving perception hardware, LiDAR and onboard cameras stand as the primary choices for numerous automakers. Nevertheless, given LiDAR's high cost and the pivotal role of the information it gathers in decision-making systems, some have suggested using millimeter-wave radar as a substitute. Traditional millimeter-wave radar, however, has a drawback: it struggles to effectively detect the height information of targets. To address this, 4D millimeter-wave radar, capable of measuring elevation angles, has been developed. Yet, LiDAR remains indispensable in autonomous driving.

How Does Millimeter-Wave Radar Function?

To assess whether millimeter-wave radar can replace LiDAR, it's crucial to first grasp how it operates. As the name implies, millimeter-wave radar emits electromagnetic waves with millimeter-range wavelengths. These beams are sent outward and reflect back upon encountering an object. By calculating the time difference between emission and reception, the radar estimates the object's approximate distance. Utilizing phase changes or frequency shifts (the Doppler effect), it can also measure the object's speed. Through multi-antenna arrays and beamforming technology, the radar determines the target's azimuth angle. In essence, millimeter-wave radar perceives targets' positions and speeds by relying on 'wave time' and 'wave frequency/phase changes.'

This mechanism renders millimeter-wave radar highly sensitive to motion, enabling it to directly output speed information. This is particularly valuable in vehicle dynamic perception, as it can directly ascertain whether an object is stationary or in motion. Owing to its strong penetration, millimeter-wave radar experiences significantly less attenuation in rain, fog, and dust compared to lasers, allowing it to stably detect echoes in adverse weather. Consequently, it's widely utilized in automotive auxiliary perception systems.

Millimeter-wave radar primarily extracts parameters like target distance, speed, and scattering characteristics from electromagnetic wave echoes. Compared to LiDAR, it struggles to provide high-density spatial geometric information. In essence, millimeter-wave radar excels at determining 'whether a target exists, its distance, and speed,' but it's relatively weak in accurately depicting objects' shapes, outlines, and details, limiting its broader applications.

Working Characteristics and Advantages of LiDAR

LiDAR employs short-pulse lasers or scanning laser beams to illuminate the surroundings, obtaining precise distances by measuring the round-trip time of light pulses ('time of flight'). Compared to millimeter waves, laser wavelengths are shorter, and the beams are more concentrated with a smaller divergence angle. Thus, LiDAR can focus energy within a smaller angular range, achieving higher angular resolution and denser point clouds. The advantage of high-density point clouds is the clear presentation of 3D geometric structures, such as pedestrian outlines, car door edges, and curb details, which are invaluable for target classification, precise positioning, and scene understanding.

LiDAR's strengths in ranging accuracy and angular resolution endow it with powerful geometric perception capabilities, enabling it to generate dense and structured 3D point clouds for precise segmentation, boundary detection, and shape inference. In static or slow-moving scenarios, LiDAR can accurately depict object shapes, crucial for high-definition map construction, localization, and fine semantic segmentation.

Of course, LiDAR has its drawbacks. Its short wavelength makes photons susceptible to scattering in rain, fog, or snow, degrading echo quality. Direct exposure to strong light may also cause saturation or false positives. Additionally, LiDAR faces challenges in cost, size, and reliability, though these issues are gradually improving with solid-state and mass production advancements. From a perception capability standpoint, LiDAR's 'spatial resolution' and 'point cloud structuration' are unmatched by millimeter-wave radar.

What Are the Limitations of Millimeter-Wave Radar?

Understanding the principles and capabilities of millimeter-wave radar makes it clearer why it can't fully replace LiDAR. The lateral (angular) resolution of millimeter-wave radar is constrained by the physical size of the antenna array and the wavelength. Achieving angular resolution as fine as LiDAR's would necessitate a very large or complex antenna array, introducing cost and power consumption challenges. While longitudinal (distance) resolution can be enhanced through techniques like spread spectrum and frequency-modulated continuous wave, millimeter-wave radar still falls short of LiDAR's dense point clouds in terms of point cloud density and shape reconstruction capabilities. Millimeter-wave radar can only inform you 'where something is and what it's doing,' but it struggles to accurately reconstruct object shapes, making it difficult to support decisions requiring fine geometric information.

Millimeter-wave radar is also highly sensitive to the electromagnetic scattering response of targets. Different materials and angles produce significantly varying reflections of millimeter waves, leading to issues like specular reflection or blind spots. Non-metallic thin objects, such as plastic sheets, fiber mesh, or suitcase edges, may appear almost 'invisible' to millimeter waves at specific angles. In contrast, lasers, with their shorter wavelengths and more concentrated energy, produce more stable echoes. While lasers generate significant scattered noise in rain or fog, creating a 'white fog' effect in point clouds, millimeter-wave radar can more reliably detect distant vehicles or obstacles in such conditions. Due to these differing sensitivities to materials and weather, simple substitution isn't feasible.

Autonomous vehicles need to know not only 'that something exists' but also 'what it is' and 'what shape it has.' Laser point clouds directly provide geometric information, enabling reliable differentiation of targets like pedestrians, bicycles, vehicles, and railings when combined with semantic algorithms. Millimeter-wave radar echoes are relatively sparse or ambiguous. While they can perform some degree of discrimination using features like micro-Doppler or echo intensity, they still fall short of LiDAR's performance in complex scenarios and near-field detail recognition. If decision-making logic relies on edge detection, contour fitting, or fine spatial segmentation, millimeter-wave radar alone struggles to provide stable and reliable results.

Autonomous driving systems necessitate redundant and explainable perception chains. LiDAR provides intuitive, easy-to-understand 3D measurements that facilitate debugging and validation. Millimeter-wave radar echo characteristics, however, require complex signal processing and algorithm interpretation. Fault modes, such as false targets caused by multipath reflections, are not easily traceable, further highlighting millimeter-wave radar's limitations. From a functional safety and regulatory compliance perspective, sensors that make judgments based on intuitive geometric information are easier to define in terms of behavioral boundaries. This is one reason why advanced autonomous driving systems still retain LiDAR as a critical perception source.

In recent years, millimeter-wave radar has been evolving toward 'imaging radar,' with new technologies like MIMO, spectrum expansion, and deep learning continuously improving its angular resolution and point cloud density. However, to fully bridge the gap with LiDAR, significant breakthroughs are needed across multiple domains, including antennas, radio frequency, bandwidth, and computing power, while also controlling costs and ensuring reliability. This isn't impossible, but full substitution is unlikely in the short term.

Final Thoughts

From the principles of millimeter-wave radar, we can see its advantages in speed measurement, robustness in adverse weather, cost, and integration ease. However, its limitations in angular resolution, detailed geometric reconstruction, and semantic discrimination prevent it from independently meeting all spatial perception needs in autonomous driving. At the forefront of intelligent driving, we believe a more stable hardware configuration path at this stage is to use a perception fusion solution. Millimeter-wave radar ensures basic safety perception in low-visibility conditions, LiDAR provides fine 3D geometric reconstruction, and visual information aids semantic understanding. Only through the collaboration of these three can issues like 'seeing clearly,' 'judging accurately,' and 'deciding whether to act' be properly addressed in complex traffic environments.

-- END --