The era of electrification has driven not only innovation in the battery industry, but also the speed of innovation in cameras, lidar, millimeter-wave radar, and other sensor devices related to autonomous driving.
From relying on radar perception, to the fusion perception of camera + radar, and then further adding lidar, these terms that we were unfamiliar with but are now familiar with are all aimed at achieving a goal of "having a higher-order automatic driving capability." Millimeter-wave radar, one of the devices for autonomous driving sensing, has also ushered in the next generation of technological evolution after years of popularization - 4D imaging radar.
So, can the 4D imaging radar bring new heights to autonomous driving?
4D imaging radar, where exactly is new
One thing we can't deny is that the current level of autonomous driving has reached the L2/L3 level, and the next development is the L4/L5 level. Therefore, the focus of the entire set of perception systems has changed from the early "control false alarm rate" to the current "avoid leakage recognition", and the market demand is promoting the progress of perception hardware.
The earliest application experience of millimeter-wave radar as we know it is ACC adaptive cruise. Until now, we can also see the figure of millimeter wave radar in cars with automatic driving functions, the detection distance is long + the penetration is strong in response to rain/snow/fog in bad weather, making it irreplaceable, and assisting high-definition cameras and lidar to achieve better perception.
Disadvantages of traditional millimeter wave radar:
1. Can not provide vertical information of the measured object, can only provide horizontal data;
2. Because there is no height information, it is difficult to identify stationary objects and cannot make accurate judgments, and relying solely on millimeter-wave radar to make decisions and judgments may lead to the risk of false braking/crashing;
3. Point cloud fusion is difficult, it is difficult to map, and it is also the weakness of millimeter-wave radar and lidar.
4D imaging radar for further technological evolution needs to meet two conditions, the first is to have longitudinal information detection capabilities, and the second is to form point cloud maps. In order to meet the above two functions, the practice within the 4D imaging radar industry is divided into two types:
1. Multi-piece millimeter wave radar transceiver MMIC cascade, representing the manufacturer ZF, with 192 channels;
2. The use of SAR technology to achieve virtual aperture function, on behalf of the manufacturer Oculii;
3. Adopt 12 transmit, 24 receive large antenna array, representing the manufacturer Huawei.
MMIC cascading is actually very easy to understand, and it is relatively simple to implement, the principle is that the MMIC transceiver uses cascading to obtain a greater gain effect than when a single transceiver is used. Another "virtual aperture" can also be called synthetic aperture (SAR) technology is relatively more interesting, this technology was previously used more commonly on space vehicles, satellites, mainly used for mapping and other purposes.
General millimeter wave radar angle resolution and radar aperture are proportional to each other, basically the city 3 round / 4 reception configuration, because the radar aperture limited angle resolution can often not be done particularly high, generally only around 10 °. Placed on the 4D imaging radar, some manufacturers use software algorithms to achieve the use of algorithms to virtualize ten times the number of antennas in the case of unchanged physical antennas, increasing the angular resolution from 10° to 1°, thus bringing more efficient identification of neighboring detected objects.
From the perspective of in-vehicle applications, this advantage is that it is not limited by the number of physical antennas (radar volume), and the virtual multiple of algorithm optimization can be increased (and there is still room for improvement). For example, Oakii's dual-chip EAGLE imaging radar uses the technology mentioned above to achieve a horizontal resolution of 0.5°, a longitudinal 1°, a detection distance of more than 350 meters, and an operating frequency of 76-81GHz.
If the algorithm is optimized again, the point cloud map formed after the angular resolution is increased is likely to reach the level of detail of lidar.
After application = high-level autonomous driving essential?
It has to be admitted that the 4D imaging radar, after inheriting the traditional millimeter wave radar has strong penetration and stable working conditions in bad weather such as rain/snow/fog, it has achieved high resolution, point cloud maps and vertical data for measuring objects, and also paved the way for the autonomous driving function to move forward to a high level.
The current mainstream autonomous driving function is still at the L2/L3 level, and the mainly dependent sensing equipment includes a combination of camera + millimeter wave radar + lidar. This approach to camera+ millimeter-wave radar is common, but it has low resolution and does not obtain enough data (e.g., altitude information); lidar can compensate for the above problems, but lidar is more expensive, limits large-scale applications, and there are unstable factors in dealing with severe weather conditions.
Then there's the question, how does 4D imaging radar push high-level driving to the ground? The answer is: yes, but in a specific scenario.
First clarify the relationship between automatic driving and imaging radar, imaging radar is only the perception of equipment hardware, the specific role is only as the perception of objects around the vehicle, data collection; the final realization of automatic driving function, or rely on algorithm + perception equipment to finally achieve.
What role can it play if it is added to an existing sensing device? Relying on good penetration, when visual perception equipment and lidar face the harsh working environment, the role of imaging radar will be amplified, so as to achieve all-weather perception (similar to lidar). It can be said that 4D imaging radar is a reinforcement to the existing perception equipment, although its function is very close to lidar, but it can not replace the existence of lidar for the time being.
In addition, there are feasible solutions for technological innovation, or the use of algorithms to do adaptive FM to increase the number of virtual antennas, for example: the expressway can be turned up to see farther, so that lidar and cameras may obtain information and plan routes in advance; on urban roads, reduce the frequency to collect close-up information and improve accuracy.
Problems to be faced before 4D imaging radar vehicles:
1. Radar using cascading methods may need to increase the physical antenna to increase power, which requires solving the problem of loading volume;
2. Radar using virtual aperture mode, due to the number of virtual antennas of about ten times, it is necessary to do a good job of antenna anti-interference problems;
3. Meet the requirements of the vehicle specification level, stable, reliable and can meet the anti-interference ability.
If the 4D imaging radar is on the car, it is likely that the advanced automatic driving scenarios that will be applied first are high-speed road section assisted driving and high-level intelligent parking functions; one is the improvement of the forward radar angle resolution, and the other is to improve the close-range detection capability. More often, 4D imaging radar may serve as a complement to lidar.
summary
4D imaging radar makes up for the high resolution that traditional millimeter-wave radar does not have, but it also has the reliability to obtain valid information and work stably in any extreme environment, which makes up for the lack of lidar. At the same time, the manufacturing technology of the major Tier 1 for millimeter-wave radar is also relatively mature, and the cost of 4D imaging radar can be well controlled, and it is more likely to be mass-produced.
At present, whether it is Tier 1 head enterprises, or start-ups, technology companies, they have entered the field of 4D imaging radar. Hardware is easier to solve than the algorithm, but the algorithm of the 4D imaging radar used behind the hardware will also become another point of competition, and with the iteration of technology, 4D imaging radar may gradually approach the use of lidar.