In conclusion, the biggest difference between 4D radar and traditional radar that can only see a two-dimensional surface is not simply the addition of one more data point, but rather its ability to provide a "more three-dimensional and reliable view." For autonomous driving, radar is like the ears and touch of a human. It hears (receives) echoes, calculates distance and speed, and then tells the brain when to brake or swerve. It can better distinguish objects at different heights on the road, providing a more structured point cloud, helping the vehicle make more reliable judgments in complex environments.
What is 4D millimeter-wave radar, and how does it differ from 3D/traditional radar?
"4D" is not a mathematical term, but rather refers to the functionality of radar from the perspective of "perceptual dimensions." Early automotive radar primarily measured distance, and incidentally, radial velocity (i.e., whether the target is approaching or moving away, and at what speed). Later, radar added antenna arrays, enabling it to determine azimuth (left and right), sometimes referred to as "3D," although the terminology for "3D/4D" is not entirely consistent across manufacturers and articles. Here, I interpret 4D radar as providing four types of information simultaneously: distance, azimuth, velocity, and altitude (which can also be understood as vertical angle). The addition of altitude transforms radar from a "two-dimensional shadow" into a three-dimensional perception with depth.
To put it simply, traditional radar is like standing on the side of the road looking through binoculars; it can only tell you how far away something is and what direction it's in. 4D radar, on the other hand, is like having a stereo camera with height measurement capabilities. It not only knows where things are, but also their height, and can distinguish the height differences between small pebbles on the ground, roadside guardrails, and people's heads. This height information is extremely valuable for scenarios involving pedestrians, cyclists, truck roofs, and road signposts.
How does it "judge the altitude"?
To enable radar to determine altitude, two key aspects are crucial: antenna layout and signal processing. Early radars had fewer antennas, with a limited number of transmitting and receiving antennas, making it impossible to accurately determine vertical direction using angular differences. Modern 4D radars utilize more antenna elements in both the transmitting and receiving sections, arranged according to specific rules. This allows the radar to measure the phase differences of signals between different antennas. By piecing together these phase differences, the radar can calculate the angles from which the target is viewed from the side and from above, thus obtaining altitude information.
Another key technology is beamforming and MIMO. MIMO combines multiple transmitters and multiple receivers, obtaining more independent measurement samples by combining the phase information of different transmitter/receiver pairs. Beamforming, like adjusting the beam direction of a flashlight, concentrates energy in a specific direction, improving resolution for a particular location. Combined with fast time-frequency coding and more powerful signal processors, radar can obtain finer angle and distance measurements within the same time frame. With the increased computing power of modern chips, these arduous mathematical calculations can be completed at the sensor end, outputting a "cleaner" point cloud to the vehicle's central control system.
To use a more easily understood analogy, imagine you want to take a 3D picture of an object. Multiple small cameras take pictures from different heights and angles, and then the pictures are combined to obtain the target's 3D shape. 4D radar does something similar, except it uses electromagnetic waves instead of light. Antenna arrays and signal processing replace multiple cameras and stitching algorithms. Electromagnetic waves have the ability to penetrate fog and rain, which is one of the reasons why radar is more reliable than cameras in inclement weather.
What improvements does 4D radar bring compared to 3D radar/traditional radar in real-world autonomous driving scenarios?
With height information and denser point clouds, the system can do more and is more reliable. For example, consider a roadside guardrail with crossbars above it. If a vehicle only knows the distance and direction, it might treat the guardrail and crossbars as a single unit, lacking precision. 4D radar can tell the system which echoes come from higher positions and which from the ground, allowing the decision-maker to more accurately determine which objects are obstacles and which are simply high-altitude markers, thus preventing obstruction. Another example is at tunnel or underpass entrances, where the remaining clearance for tall trucks differs from that of ordinary cars. 4D radar can help determine the presence of high-roofed obstacles, prompting the vehicle to slow down or change lanes when necessary.
In pedestrian and cyclist identification, 4D radar provides more stable geometric information. While cameras are prone to malfunction at night or in backlight, and lidar is affected by rain and snow, radar is more forgiving of these conditions. When camera information is unreliable, the altitude and speed information provided by radar serves as an important backup, preventing incorrect overtaking or misjudging distances. By combining vehicle speed, orientation, and altitude, tracking dynamic targets is less likely to confuse two people or one person with an object, thus reducing misjudgments.
Another crucial advantage of 4D radar is its more stable motion prediction. Vehicles need to predict the trajectory of targets ahead to determine their next move. Precise angle and speed information means that tracking algorithms can more accurately maintain the target's trajectory and are less likely to "lose the target" in short bursts of noise. This makes path planning and decision-making smoother, reducing frequent braking or unnecessary evasive maneuvers caused by perceived vibrations.
4D radar is not a panacea; what are its limitations and practical trade-offs?
While 4D radar sounds incredibly powerful, it still has some unavoidable drawbacks. First, the point cloud density and detail of radar still cannot match that of high-line-count LiDAR. LiDAR provides clearer images of long-range and small target outlines, and even 4D radar cannot completely replace the full functionality of high-end LiDAR. Second, height resolution has physical limitations. The number of antennas, the vehicle's mounting location, and signal bandwidth all affect the vertical precision, so height information may not be very accurate in certain scenarios. Third, it increases cost and engineering workload. More antennas and more complex signal processing mean the sensor itself is more expensive, and electromagnetic compatibility, heat dissipation, and housing design must be more careful to meet automotive-grade requirements. Fourth, data processing is crucial for usability. The point cloud output by 4D radar requires corresponding algorithms and specialized labeled datasets to train the model. Without mature algorithms, rich point clouds may become a "noise pile," unable to be directly converted into reliable decisions.
This is why many manufacturers don't treat 4D radar as a standalone "panacea." A more realistic approach is to use it as an important technological complement, working in conjunction with cameras and LiDAR. Cameras handle high-resolution semantic understanding and color information, LiDAR provides dense 3D shape data, and 4D radar provides stable distance, velocity, and altitude information in low-visibility or complex electromagnetic environments. The three sensors working together are more reliable than any one of them operating alone.
How can I use 4D radar effectively and get the best value for my money?
To truly realize the value of 4D radar in vehicles, several things must be considered. The sensor installation location must be repeatedly verified, as its position determines the field of view, obstruction, and antenna efficiency; haphazard placement will significantly degrade performance. Time synchronization and spatial calibration are also crucial; data from radar, cameras, and LiDAR must be aligned in time and calibrated in space for effective fusion. At the algorithm level, lightweight online processing should be implemented first, placing judgments with high uncertainty in conservative strategies, and then delegating complex semantic reasoning to central computing or offline model optimization. Finally, reliability during mass production must be considered; automotive-grade packaging, vibration resistance, high and low temperature resistance, and electromagnetic compatibility are all essential hurdles to overcome.
From a product selection perspective, automakers need to align cost, expected functionality, and target scenarios. Not every vehicle requires state-of-the-art 4D radar. For some shared mobility vehicles primarily used in low-speed urban environments, 4D radar can provide a significant safety improvement; while for mass-produced family cars with limited budgets and simpler scenarios, a lower-cost sensor combination and more software optimizations may be chosen.
How will 4D radar impact the future of autonomous driving?
With technological advancements, chips will become cheaper, antenna manufacturing will mature, and algorithms will be able to better interpret radar point clouds. In the future, 4D radar will become increasingly common, especially in scenarios requiring reliable all-weather, all-time sensing, where it may become a primary hardware component. However, it won't completely replace LiDAR in the short term, but it will make sensor fusion more robust, effectively reducing the risk of single-point failure of optical sensors in adverse weather conditions.
If you think of autonomous driving as a football team, cameras are the forwards responsible for seeing details, lidar is the midfielder responsible for building a detailed 3D scene, and 4D radar is the defender who intercepts the opponent's attack at crucial moments; in rain or fog, it is often the most reliable defensive line. Only by properly assembling the team members can you play a good game. The emergence of 4D radar allows this team to maintain its rhythm even in wind and rain, which is its most direct and practical value.
