said very seriously that Chinese car companies are really leading the trend in the field of new energy smart cars.
In the automotive field, if we have been chasing after many years, then the situation may have undergone some changes now. Chinese automobile brands have gradually surpassed and led in some areas.
pure visual school or lidar school, who has a greater chance of winning?
The field of autonomous driving can be divided into two major technical schools: one is the purely visual school represented by Tesla , and the other is lidar represented by NIO , Xpeng and Google . group.
Why is Tesla so unique? Musk gave several reasons: First, multiple sensors working together sound good, but there will be conflicting sensing results, which is difficult to handle. Second, lidar cannot be used alone. It also requires high-precision map matching, and the technology is not yet mature.
Of course, there is another reason that Musk did not say, which is that lidar is very expensive, and an car camera costs only 30 US dollars.
Of course, pure visual systems also have many technical obstacles, such as inaccurate distance measurement, severe distortion when backlit, etc. Therefore, the opportunity for lidar comes.
At night when there are no street lights or any lighting, because the radar actively emits laser detection, the vehicle can still maintain its judgment when there is sufficient light.
Although ultrasonic radar and millimeter wave radar have similar functions, the resolution of three-dimensional images they form is far less than that of laser, so relying only on one lidar and a simple principle of "speed × time = distance" can Can complete the task of "perception" very well.
Of course, lidar can only determine the three-dimensional outline of an object, and cannot identify content such as color, texture, and material in a very detailed manner. Therefore, cars equipped with lidar technology are also equipped with cameras to collect color and texture information of objects, such as traffic lights, traffic signs, etc. Anything that lidar is not very sensitive to is supplemented by cameras.
So to summarize, one is the visual school represented by Tesla, which adopts a low-perception + high-processing approach, and the hardware cost is relatively low. The other is the lidar school represented by NIO and other car companies, which uses high perception + high processing power, but the hardware cost is relatively high.
How powerful is the lidar of NIO ET7?
On March 28, 2022, NIO’s smart flagship electric sedan ET7 started user delivery. According to NIO officials, NIO ET7 has become the world’s first batch-delivered ultra-long-range, high-precision lidar model.
Next, let’s take a look at the strength of NIO ET7’s lidar? Compared with NIO's previous products, will NIO's ET7, which has lidar, reach a higher level in terms of autonomous driving?
Weilai ET7 is equipped with ultra-long-range and high-precision lidar from Innovusion, with a maximum detection range of 500 meters, a detection range of 250 meters under with 10% reflectivity, and a 120° ultra-wide level With a viewing angle and 0.06°*0.06° ultra-high resolution, it is the world’s first 1550nm lidar to achieve mass production.
Of course, what needs to be explained here is that the lidar of Weilai ET7 is still a semi-solid state.
There are currently three technical routes for lidar: mechanical, semi-solid and solid-state. Mechanical lidar was the first to be mass-produced. However, it has several natural flaws that are difficult to overcome: first, it is driven by a mechanical device, which makes the product larger in size and unsightly when mounted on the car; second, it is mechanically driven. The system is complex and precise, prone to failure, not safe enough, and has a short service life. Third, single-piece production requires manual adjustment, which makes mass production difficult and costly.
Solid-state lidar is just the opposite. It can be integrated with the chip to become a pure electronic product. It can be made small in size, reliable and durable, and does not require manual adjustment. As long as the supply chain is in place, mass production will be easy and costs will decrease.However, its technological content is higher, so engineering tuning takes longer. Although the industry is optimistic, solid-state lidar will take at least three or four years to achieve mass production.
The semi-solid lidar of NIO ET7 is already the best solution at present.
In addition, semi-solid-state lidar also has two product branches, 905nm and 1550nm. These two numbers describe the wavelength of the laser. The laser is essentially electromagnetic wave . Different wavelengths have different effects on the human eye. 905nm is just within the absorbable range of the human eye. If the emission power is too high, it will cause damage to the retina of the human eye.
Therefore, in order to protect the vision of pedestrians outside the car, the 905nm lidar has to reduce part of its transmission power. If the power is low, the performance will decrease and the range will not be far. If you encounter weather with low visibility such as rain, snow, fog, etc., the 905nm lidar will not work properly.
The 1550nm lidar chosen by Weilai ET7 emits an electromagnetic wave that is safe for human eyes, so the transmission power can be increased with confidence, and the detection range will be longer.
Weilai ET7 lidar has a maximum detection range of up to 500 meters, and the detection distance is under the 10% reflectivity standard (it should be noted that 10% reflectivity is very, very rare in nature, and the vast majority of reflections The rate is higher than 10%), up to 250 meters, leading the industry. The farther the detection distance is, the more time and distance will be available to stop the vehicle in dangerous situations to ensure driving safety.
In addition to seeing far, the rigid requirement for lidar is to see more clearly. The
ET7 lidar has a "fixed gaze" function and a bionic human eye that can flexibly zoom and produce high-density point cloud with an angular resolution of up to 0.06°*0.06° in the key driving visual field, allowing targets in this area to be seen more clearly. The fixed gaze function range of ET7 lidar is 25°H (horizontal) * 9.6°V (vertical). At 50m in front, the coverage of this area: horizontal coverage reaches 10 lanes, vertical coverage is greater than the height of one floor), which can fully cover the driving area ahead.
ET7 LiDAR focuses on the area and strategically chooses to encrypt key areas instead of covering the entire area. It can increase the resolution of any area at any time as needed. Avoid occupying unnecessary computing resources, increase power consumption, bandwidth, etc., improve the effectiveness of data utilization, reduce algorithm pressure, and be more conducive to the operation of the vehicle sensing system, truly realizing "using the "point" on the blade".
This is like the fixed gaze function of the human eye. Our human eyes are very magical sensors, and the function of fixed gaze is very magical. When we want to see an object clearly, we can stare at it, stare at it, and we can see it very clearly. In fact, the principle behind it is to focus all the photosensitive pixels of the human eye into such a small range for him.
ROI is dynamically adjustable to achieve better perception and tracking of vehicles and pedestrians. It can effectively detect distant dangerous targets in advance even in working conditions such as large curvature turns or up and down slopes ahead. Only by discovering earlier can we decelerate earlier and issue warnings to users earlier, thus improving the reliability and safety of autonomous driving.
Moreover, for lidar, seeing steadily is also a necessary quality.
is considered stable in the industry, mainly relying on POD, Probability of Detection, which is generally the number of laser beams emitted for more than 100 consecutive frames (i.e., the number of theoretical points) and the number of detected laser beams (i.e., the number of effective points) ratio. POD reflects the ability and stability of lidar to receive return points and is an important indicator of lidar performance.
is just like the housing availability rate of and . For houses with the same building area, of course, the house with the higher housing availability rate will be more comfortable to live in. The same is true on lidar.
The lidar equipped on NIO has excellent detection capabilities. The detection probability of detecting objects with 10% reflectivity at 250m exceeds 90%, leading the industry.A higher POD can allow the vehicle to perceive the target object more clearly, reduce the pressure on the algorithm, achieve a longer effective perception distance, and improve the overall safety of the autonomous driving system .
As the first mass-produced ultra-long-range lidar, it can indeed make up for the natural flaws of cameras at the perceptual level. More importantly, NIO’s use of this lidar has particularly obvious advantages in performance, and it has also overcome Despite the cost problem, the realization of mass production and installation is a qualitative leap for the development of China's autonomous driving technology.
