Ideal car this wave of marketing operations, has been "squeezing toothpaste", "hanging appetite" these ways to a certain height, squeezed for a few weeks, and finally squeezed out the attention of "intelligent driving" related content, this time spoilers, but also let the ideal fully self-developed flagship intelligent driving system - ideal AD Max, officially met with us, but also let us see the ideal L9 vision sensor "eye" in the Wangjing.
What's the secret on the roof of the ideal L9 "obtrusive"?
In the previous spoiler photos, we can already clearly see the "towering" lidar above the ideal L9, this model of AT128 lidar from the domestic lidar field upstart - Hesai Technology, Ideal Auto and Hesai Technology reached a cooperation in July last year, this time equipped with Hesai for high-level assisted driving to create a 128-line hybrid solid-state lidar, it is also a set, high-speed and sudden amplification of Lotus lidar suppliers.
The laser wavelength of the AT128 is 905 nm, which is lower than the 1550 nm power of the Tuduitone Falcon carried by the NIO ET7, and the detection distance will be closer, but the maximum detection distance at 10% reflectivity can also reach 200 meters, leaving sufficient redundancy space for feedback such as braking, and the field viewing angle is 120° x 25.4°, which can be understood as the detection angle of the horizontal and vertical direction of the lidar, and the general situation is symmetrical. Lidar forms a cone-shaped area based on two field of view angles to create a point cloud image.
The highlight of the ideal lidar is the use of 128 laser emitters, vertically arranged, forming a laser line, and then through a 1-dimensional rotating mirror, the line into a surface, to achieve the detection of the forward environment. This type of hybrid solid-state lidar with a rotor can obtain higher point cloud density and higher resolution in a short period of time, but the relative cost is higher, and there are bearings with fixed rotors inside, resulting in a certain impact on its service life. The Xiaopeng P5, which uses double-wedge-shaped prism lidar, does not have such a problem, and its internal wedge-shaped prism is fixed, more stable and less costly, but relatively lower in point cloud density and resolution at the same time. It depends on which part of the car company is to take into account, the cost is high but the effect is better, the cost is low, the effect is poor but the number can be increased to increase the redundancy, and the corresponding requirements for the algorithm platform are higher.
"Wei Xiaoli" invariably put the material together
With the ideal L9 announced that lidar has become its intelligent driving hardware configuration, the domestic new force "Wei Xiaoli" three also entered the "lidar" era together, the three are like saying good, know that the ideal car every Wednesday to break new material, Xiaopeng and Weilai also invariably released two lidar "show muscle" video on the front and back feet, the three lidar distribution methods have two: roof and headlight group below.
The lidar of the ideal L9 and NIO ET7 is "long" at the top of the car, and the advantages and disadvantages of this layout are very clear:
The advantages are: first of all, its layout position is high, especially the SUV model, it is not easy to be affected by occlusion, "stand high, see far", its perspective tends to be the driver's visual position, can better start from the driver's line of sight; secondly, the front collision of the vehicle will hardly damage it, the damage may be small, reducing the cost of vehicle maintenance; secondly, because it is not arranged in the front surround or bumper part, it can leave room for the camera, ultrasonic radar and millimeter wave radar, what? You said the ideal L9 didn't say there was millimeter wave radar? forehead.... Finally, this is my doubt: if when the mold was first designed, it was not intended to install lidar, and the later lidar was forced to "get on the horse" for other reasons, then where do you say it can be installed? Inside the car? The car is not good, the line of sight is blocked by the front cabin, and the top of the car must leave a position for the camera, which can only be placed on the roof of the car, and it is not coordinated, at least "there is".
Disadvantages can correspond to each advantage: since the design is on the roof, then the overall design of the car, sensory completion will be affected; at the same time, the owner's maintenance cost is low, then it means that the late manufacturers make less money, now more and more car companies design the vehicle as a vulnerable part, how much money can be made by selling the car, repairing the car is not what the manufacturer said, saying that how much money to repair for you is how much money, the car company designs it so solidly, that is, it does not consider the aftermarket at all Could it be that I'm shallow? As long as the front face does not hit the lidar will not be bad, maybe the roof is more likely to be broken, the black roof temperature under the summer sun exposure can even exceed 70 degrees, and the limit working temperature of the lidar is mostly at 80 degrees, so the long-term critical to the working conditions of the limit working temperature is bound to affect the life and effect of the lidar.
On the P5 model, the lidar is placed in a lower position under the headlight group, which is also the layout of most models in the industry.
Which layout is better depends on the basic performance of lidar, the effect of algorithm recognition, cost considerations and the degree of cooperation with other sensors, the era of vehicle-mounted lidar has just begun, let the bullet fly a little longer.
Pixels and the number of increased cameras, can really perceive 550 meters?
After talking about lidar, the perception sensor configuration of the ideal L9 is also rich, including 6 8-megapixel cameras and 5 2-megapixel cameras.
On the ideal L9, two forward 8-megapixel cameras can achieve 120-degree wide-angle coverage, and the official said that it also takes into account the identification of vehicles, pedestrians and cone barrels within the farthest 550 meters.
Once Tesla and Weilai used the forward three-eye camera, the detection distance of each camera is different, the front view main camera detection distance is generally 150 - 170m, the front view wide angle camera detection distance is about 50 meters, and the front view narrow angle camera can reach a distance of 250 m or even farther. The pixel of the ideal L9 is improved, but the detection distance is doubled directly by the camera, is it not exaggerated?
I think that detection distances of more than 500 meters are possible, but only for vehicles, and I think it is a bit exaggerated to say that pedestrians and cone barrels can also be recognized.
The 8-megapixel camera can detect pedestrians in the range of 100 - 150 meters, and in the case of a narrow viewing angle of field mode, it can detect dynamic vehicles of about 500 meters, small targets of about 180 meters.
The 8-megapixel camera MFC535 released by Continental Germany, the field of view of this camera is 125 °, and can recognize 100m traffic signs, 140m bicycles, pedestrians, 160m motorcycles and tricycles, 500m cars; for the 8 million pixel camera of the vehicle specification level, the current domestic Niolai ET7, 2021 ideal ONE Krypton 001 and other models are in use. The NIO ET7 is equipped with 11 8-megapixel HIGH-definition cameras from Joyson Electronics, which can detect vehicles of about 687m, cone barrels of about 262m, and people of about 223m at the farthest.
From this, we can clearly see that although the current camera pixel is very high, the basic performance has been greatly improved, but in the range of more than 500 meters for object recognition or stay on large objects such as vehicles, for pedestrians, cone barrels and so on still stay at about 200 meters.
In addition, the ideal L9 has two 8MP side front cameras, two 8 million side front cameras and 1 x 2 million rear camera to complete the 360-degree perception of the vehicle's body. The ideal use of 4 2 million pixel cameras, compared to the current model has a certain improvement, due to the positive rear perception demand is small, the ideal only to install a 2 million pixel camera, the perception ability of the rear camera is generally about 50 meters, and basically act on the auxiliary parking, 2 million pixels can also be said, than the Toyota Lexus camera is clear on the line, this does not have to be too entangled in this aspect, but a thought of this is the flagship product of the ideal car, and the price is close to 500,000, That should still be the best lens, Weilai in this regard to do a good job, its own flagship car ET7 directly with 8 million pixels lens piled up the whole car, the effect we said otherwise, but this is indeed the flagship should look.
For millimeter-wave radar, the ideal is "as secretive as deep"
As mentioned above, the ideal did not announce whether there is a millimeter wave radar hardware configuration on the L9, on the one hand, it is possible to continue to carry 5 millimeter wave radars like the ideal ONE, but the chip is tight, at that time, the ideal ONE of 5 millimeter wave radars was equipped with 3, and the flagship model may also face this problem.
The other is to eliminate the need for millimeter-wave radar, and use perception cameras and lidar to solve the problem together, not only without worrying about chip shortages, but also reducing costs. The ideal L9's lidar can only cover the 120-degree angle ahead, and without millimeter-wave radar assistance, the vehicle's remaining 240-degree angle is only the camera to determine the object and the distance to the object.
At present, visual algorithms and fusion algorithms are the existence of their own advantages and disadvantages, the advantage of a single visual algorithm is that the decision-making layer only needs to receive the camera's information, but the visual algorithm is not completely rational, such as the ghost brake phenomenon we often say, visual algorithms mistakenly take shadows or other unreal things as objects caused. Although the fusion algorithm has millimeter-wave radar for accurate ranging, there will also be a phenomenon that the camera and the millimeter-wave radar measurement data are not uniform, and it is difficult to judge when two different data are given to the decision-making level.
The ideal should not forget that there were major hidden dangers in the automatic lane change due to the lack of lateral millimeter wave radar.
In terms of software systems, the intelligent driving computing power platform contains two NVIDIA Orin-X processors, with a final force of 508Tops, and the dual processors are redundant with each other's computing power, making the intelligent driving system run more stable. Compared with the ideal ONE with only 10Tops, the computing power is very different, the number of cameras has increased from 5 to 11, the amount of data has increased significantly, and the algorithm platform with greater redundancy has also laid the foundation for future functional innovation and further improvement of the system.
summary:
At the level of intelligent driving, the hardware performance of the ideal L9 is relatively decent, but the valuable thing is that the whole set of intelligent driving system is fully self-developed, and the actual performance depends on the automatic driving algorithm, data, perception fusion and other aspects of technology accumulation, the ideal is still growing rapidly, compared to Weilai and Xiaopeng, the ideal NOA is more like a catch-up. Such a set of software and hardware has been almost completely displayed to us, so I predict that next week the ideal will "squeeze out" the detailed functions of intelligent driving, as well as OTAs, etc., and then analyze with you whether these functions of the ideal L9 are outstanding.