Four automotive CEOs held a roundtable discussion online about where and how many lidars should be installed.
Let's take a look at the different views expressed by each of the four people:
Ideal Car CEO Li thought: Put one on the roof, and put two on the cover or bumper, there is no difference in performance, and even the performance of a single one on the top of the head will be better. However, in terms of pedestrian collisions, repair costs, and vibration control (linking the main body), the roof is optimal. The only problem is that the shape of the roof lidar will need to be adapted, because it is too much like a mobile suit - Gundam.
Xia Yiping, CEO of Jidu Automobile: There is still a difference, the 120-degree FOV and the 180-degree FOV are still not the same, and the resolved kernel case is also different, so there is still a difference in the ability and experience of the product and even the safety.
He Xiaopeng, CEO of Xiaopeng Motors: Two lidars are definitely far better than one in terms of performance, and the difference can be seen in the future in urban NGP and robustness. In fact, the hardware configuration simply represents a possibility of potential, ultimately a balance of safety, cost, scenario and capability, still the barrel effect rather than sprint logic.
Shen Hui, CEO of WM Motors: Regardless of the number or location, or depending on the actual use environment, the WM M7 has a horizontal viewing angle coverage of 330° through the layout of 1 + 2 on both sides of the head, which can better take care of the corner case in most cases. In this way, perhaps 3 is the optimal solution for the current level of perception and safety balance, too few blind spots, too much meaningless.
It can be seen that behind the participation in this roundtable discussion, the models launched by each company are proper "lidar players", and the fuse of this roundtable discussion is that Jidu Automobile revealed that the first model, SIMU Car, will install 2 laser radars on the front cover of the vehicle.
On April 18, Jidu Automobile released a preview of Jidu's first model, the SIMU Car, which only focused on the front cover of the vehicle equipped with two retractable lidars.
There has been controversy about the loading location of lidar, including the roof faction led by the NIO ET7, ideal L9, and Zhiji L7; the bumper faction led by the Polar Arc S and Xiaopeng G9, and the Jidu SIMU Car, which was the first to install lidar on the front cover, was its own faction.
So, where is the lidar most reasonable, and which school of view is more argumentative?
01
God's perspective, manhole cover's perspective, pedestrian's perspective?
In fact, if we talk about vehicle-mounted lidar earlier, most people's impression of it must be a large flower pot on the roof of the unmanned test car, which is one of the typical forms of mechanical products.
However, given that car companies will never sell a consumer product with a "big steel pot overhead", then their special requirements for vehicle-grade lidar radar have become intuitive and well understood:
Volume, stability, mass production, cost.
Later, it began to develop into the mass production of hybrid solid-state lidar, Xiaopeng G9, NIO ET7, Ideal L9 and so on
Models equipped with lidar on the market are representatives of this category.
The most intuitive difference between the two is that the field of view is poor, the mechanical lidar can be rotated, and the general single layout can meet the scanning needs of all-round 360°.
Hybrid solid-state lidar cannot be rotated, and the detection angle is generally only 120°, so if there is no reasonable layout, there is a risk of blind spot viewing angle.
Therefore, regarding the layout of this kind of hybrid solid-state lidar, car companies have begun to play freely.
Ideal L9
Scenario 1: Roof (God's Perspective)
Stand tall, look far, see widely, and have a high degree of safety.
This is the logic of car companies choosing to install lidar on the roof. At present, there are not many players who recognize this plan, including Zhiji, Weima, Nezha, and SAIC Feifan, all believe that installing lidar on the roof of the car is the best solution at this stage.
First of all, the logic behind the placement of lidar on the roof of the car is actually the same as that of the sentry building, the higher the height, the more efficient it is to prevent the enemy, that is, the car company wants to achieve the effect of "looking far".
At present, the mass-produced lidar in the case of 10% reflectivity target, the general maximum detection distance of 100 meters -150 meters, at this time, if the same lidar is placed at different levels (roof, front cover, bumper) height, in the face of complex road conditions, it is obvious that the higher the position, the more conducive to monitoring the farther road conditions, in advance to make a good prediction.
For example, in the actual application of lidar, when the installation position reaches a certain height, it can be detected by crossing the front car.
The second is "wide-sighted", at this stage, the detection angle of hybrid solid-state radar used in mass production models equipped with lidar is generally only 120°, and if you want to achieve 180° in the forward direction of the vehicle, or a full-range 360° scanning, some car companies will choose dual/four lidar, or other hardware perception to make up for the remaining 60°/240° perspective.
WM M7
For example, in addition to installing a lidar on the roof, the WM M7 also installed two 8-megapixel cameras on both sides of the roof lidar, and two more lidars in the position of the front fenders on both sides of the body, with a cumulative scanning angle of 360°.
Another part of the car companies use the highest position of the vehicle to maximize the effectiveness of a single lidar in the 120° sector area, which is the same as the logical thinking of the higher the farther away.
When car companies choose to install lidar on the roof, they will also take into account the position of the lower edge of the viewing angle and the roof cover, and generally adjust the downward tilt of several degrees according to the different models.
NIO ET7
Finally, the safety factor, from an engineering point of view, the lidar installed on the roof of the car, can avoid the occurrence of mud, flying stones, scratching and other sudden events, compared to the front cover, bumper position, the probability of the roof is the smallest; and including pedestrian safety collision and better vibration control.
On the contrary, the placement of lidar on the roof of the car will also have great controversy about the appearance of the vehicle, such as Li wants to describe its shape as a mech warrior - Gundam, and He Xiaopeng describes this shape with "rabbit rake ears".
At the same time, it is also necessary to consider the existence of certain blind spots in near-field low obstacles, as well as unavoidable factors such as wind resistance coefficient, wind noise, exposure to sunlight, and heat dissipation.
Solution 2: Front bumper (manhole cover view)
Take into account the aesthetics of modeling and ensure the blind spot of low obstacles in the near field.
This method almost perfectly conceals the lidar, and if the front bumper is blackened, it can basically meet the traditional aesthetics of consumers.
At the same time, the lidar is placed in the bumper position, the blind area of vision in the vertical direction is smaller, and the identification of low obstacles in the near field is better.
For example, when encountering complex scenes such as entering and leaving the basement, rushing to congestion, and crossing intersections, it is more effective to detect the perspective of lateral vehicles and pedestrians.
However, due to the low installation position, there must be certain drawbacks for the field of view scanning range of lidar, and it is inevitable that there will be occlusion and blind spots.
Zhang Lichen, brand director of Tanwei Technology, said that the main function of lidar installed under the headlights is to supplement or assist obstacle avoidance, but there is an unavoidable blind spot problem in the position below.
The position above can be a perception function or a positioning function, but the positioning function requires 360° information, and a single semi-solid-state lidar must not be able to achieve the positioning function.
Therefore, players who use this set of schemes will basically choose multiple lidar for safety redundancy, including polar arc alpha S HI version, Xiaopeng P5/G9, and Avita 11.
In addition to the rise in hardware costs, maintenance costs are also an inevitable factor.
For example, collisions, scraping, flying stones, etc. all have the probability of causing damage to the lidar placed at the bumper, and this concept is not low in the daily driving process.
Option 3: Canopy (pedestrian view)
As the pioneer of this program, Xia Yiping believes that this scheme is extremely innovative, and the location of lidar is not only distinctive, the appearance is very plus, and it can also play a better advantage for "ghost probes".
It is true that from an engineering point of view, the position of the front cover is obviously better than the layout of the bumper, but in practical application scenarios, there is one of the biggest controversial issues in the scheme of this cabin cover - pedestrian collision protection.
This is also the reason why Li wants to think that this solution is the most unlikely to be realized, li wants to believe that lidar is not suitable for the bonnet, because this model cannot pass the latest pedestrian collision regulations.
In the China Automobile Research Institute test, there is a pedestrian protection and active safety, which tests the degree of damage to pedestrians in areas such as the bumper, engine canopy, front windshield and A-pillar of the vehicle itself after colliding with pedestrians. In other words, the test is whether the structural design and materials of the vehicle are friendly enough for vulnerable groups on the road.
For example, the Xiaopeng P7, in the test link of pedestrian protection, is on the edge of the passing line with a score of 9.767 out of 15 points.
The placement of lidar on both sides of the hatch, and do not consider whether it can pass the collision regulations, is also a very big hidden danger for pedestrian protection and other issues, and the high probability of test results will refresh the bottom line.
In addition, even if the "clamshell" design is used to assist, it is necessary to consider the material of the cabin cover, the offset of the installation angle caused by daily driving vibrations, and the maintenance cost.
Therefore, compared with the previous two sets of more "mature" solutions, the cabin canopy plan has many problems that cannot be avoided.
Scheme 4: Clamshell type, headlights, built-in front windshield
Since the front of the car and the hatch scheme have an impact on the aesthetics of the vehicle's appearance, and the maintenance cost of the bumper is too high, why not directly embed the lidar?
In fact, wherever you can think of in the car, the car companies have basically applied for patents one by one.
For example, the retractable design is also gradually recognized by car companies as one of the solutions, and this program is actually an upgraded version of the roof placement of lidar, such as Lotus ELETRE to adopt this program.
Previously, Ideal also filed a patent similar to a retractable lidar component.
This patent is actually a driving mechanism at the same time as the layout of the lidar, and its driving mechanism is used to drive the radar body to flip between the first and second positions, similar to the scheme of the concentration SIMU Car.
The simple understanding is that when the radar body does not need to be used, the drive mechanism drives the radar body set at the opening of the roof outer plate to flip, so that the radar body reaches down through the opening into the space limited by the roof outer plate and the roof inner plate.
Another patented position is the built-in front windshield.
Apple has filed a patent application for a device for on-board lidar equipment, which is mounted in the same way as Tesla's front trinocular camera, which also installs lidar behind the windshield of the vehicle.
This solution, whether it is the appearance shape, layout position, or maintenance cost, is the optimal solution, but the signal attenuation problem caused by glass will not be solved for a while and a half.
Compared with the camera, the windshield of the curved structure will cause the attenuation of the lidar electromagnetic wave signal, and the light transmittance of the windshield under the provisions of the "Automotive Safety Glass" is only not less than 70%, which is not a high-transmitting glass that meets the needs of lidar, and the transmittance of high-transmittance glass is required to reach more than 90%, which will bring about an increase in cost. At present, it is still difficult.
Of course, the layout of the lidar position has also tried the solution of inline headlights.
Lidar company Cepton has announced a partnership with Koito, Japan's largest automotive lighting company, which will provide the latter with a customized micro-lidar solution to install lidar into the headlights.
The two companies plan to start with ADAS and extend to more advanced autonomous driving, and the results of the first phase of cooperation will be announced this year.
Xia Yiping also replied to the heart of the car: in the future, the integration and integration of lidar and lamps is definitely a trend.
At present, there are six solutions for the loading position layout of vehicle-mounted lidar from the roof to the embedded headlights, each with different advantages and disadvantages.
Industry insiders said that there is no only answer to where lidar is installed, and it is still necessary to combine customer-defined autonomous driving function scenarios to do the optimal design, and mass production is a double test of technology + engineering.
Cui Dixiao, chief scientist of Zhijia Technology, said that in what position, specifically to see what functions to do, if it is to be perceived, it is more appropriate to place it on both sides of the headlights. If it is to do positioning, it will be better to put it on the top of the front gear, and the lidar placed on the top will see higher and farther.
"This is a relative concept that depends on the needs of the algorithm, for example, perceptual algorithms are generally more concerned with dynamic traffic participants on the road surface, some positioning algorithms are more concerned with buildings outside the road surface, and the requirements of the two types of algorithms can be met separately by different installations of lasers." Cui Dixiao added.
So, since the position layout of lidar depends on the needs of different algorithms, is the number of loads only redundant?
02
One guarantee, four guarantees?
As the second topic of this roundtable forum, Li Xiang's point of view is that thanks to the advantageous position of the roof, there is no difference between single and double, and even the performance is better than the double layout.
Xia Yiping and He Xiaopeng's view is that the more the safer, while Shen Hui is more neutral, looking at cost in the short term and technology in the long run.
This topic has always been highly controversial, including last year's salon mecha dragon official announcement equipped with 4 Huawei 96-line lidar, for a time became the most equipped with lidar on the market model, compared to a single lidar, not to mention the comprehensive parameters of the crushing, the number of simple lidar has made the mech dragon momentum for a long time.
In fact, the number of lidar depends on the range of perception and scene capabilities that need to be met.
For example, the perceptual range, the detection angle of hybrid solid-state lidar is generally only 120°:
Some car companies believe that this is enough to achieve the function of high-end automatic driving;
Some car companies believe that two 120° superpositions are needed to talk about "safety";
Some car companies believe that four of them can form 360 ° to achieve high-level automatic driving.
As for the scene function, taking the three major scenarios of city, high-speed and parking as an example, it can be divided into:
L2 has been able to help people on structured roads and highways;
L3's vision is structured roads that replace people at low speed and help people at high speeds;
L4 is driving autonomously at full speed on structured roads.
In the urban limited area of automatic driving, the three levels also need to derive countless detailed scenes.
Cui Dixiao said that because of the principle characteristics of lidar itself, it cannot do all its functions.
Then a single lidar cannot achieve multiple functions in the process of use, and only through multiple lidar can the functions required by the autonomous vehicle be covered. This is why high-end autonomous driving functions need to be equipped with so many lidar, and some even pile up four lidars.
In theory, the more the number of lidar, the higher the degree of intelligence of the vehicle, but there is a golden balance point between the actual degree of vehicle intelligence and the number of sensors, blindly stacking the number, ignoring the computing power of the vehicle itself, and may be overloaded like a smart phone and there will be caton, downtime and so on.
Of course, the number of lidar, in any case, can not avoid the cost of consideration.
With the development of recent years, the domestic hybrid solid-state lidar quotation is generally pressed to less than 8,000 yuan, such as the First Domestic LiDAR model delivered Xiaopeng P5, after-sales replacement price of 6916 yuan / piece, then the price of two is already close to 15,000 yuan.
What is the concept of accessories at this price for a car?
Toyota CVT gearbox in the domestic installed capacity has exceeded 10 million units, the technology is quite mature, the domestic aftermarket price steel strip version of the lowest price in about 17,000 yuan.
As one of the small branches of the autonomous driving perception layer, lidar has almost caught up with the price of one of the three major parts of traditional cars.
Although consumers are still willing to pay a huge cost for lidar, it is still unknown whether each lidar can "make the most of it" or pay for the marketing of enterprises.
Huawei's 96-line lidar developed at the end of 2020 has reached the scale of 100,000 sets per line per year, and the future plan will reduce the cost to $200 or even $100, which is enough as a very reasonable reference price.
03
When will you be able to come out and speak with a real knife and gun
At present, the models planned to be delivered during the year include NIO ET7, NIO ET5, Polar Arc Alpha S HI Edition, Xiaopeng G9 and so on.
But this is only the timeline of delivery, and does not mean that the software services corresponding to lidar have been able to be enjoyed.
Taking Wei Xiaoli as an example, all three have encountered a very difficult problem:
NIO ET7 is equipped with NAD advanced driver assistance system, which first opens the auxiliary driving system, and then opens NAD services in some areas;
The ideal L9 is equipped with the AD Max advanced driver assistance system, and the relevant functions depend on administrative approval and will be gradually turned on in various cities;
The Xiaopeng G9 is equipped with the XPILOT 4.0 intelligent driver assistance system, which will be launched in 2023 and will be opened in various cities under the premise of regulations.
It is understood that the latest driver assistance software systems need to be adapted to their corresponding lidar hardware, otherwise lidar cannot play a role at all.
In other words, the hardware gets in the car first, and the software depends on the situation.
The three head players are still like this, Zhiji L7, Weima M7, Feifan R7, Avita 11, Mecha Dragon, Nezha S and other models equipped with lidar, whether they can immediately enjoy the corresponding software services as soon as they are listed is still unknown.
In addition, even if the most ideal time is for consumers to enjoy the product experience brought by lidar in 2023, there are still a bunch of real problems behind the actual effect and daily maintenance.
Industry insiders said that there are still many difficulties in the process of lidar hardware evaluation by car companies, and there are great differences in the test methods and test indicators of different car companies, and the test conclusions drawn are also very different.
When many lidar companies first get on the car, various parameters such as point frequency, accuracy, detection distance and so on are perfect, but lidar works in the environment of "wind and sun, turbulent and displaced" all year round, and the effect is greatly reduced after half a year.
Any mistake in any small link of lidar will affect the work of the entire automatic driving system, and this also involves the problem of the lidar after-sales system.
Taking the head of the new power player as an example, this question currently produces two kinds of replies:
There is no policy in this regard;
Replacement or maintenance, depending on the specific situation of the specific analysis, there is no relative standard process.
In other words, these car companies have not finalized a standard and feasible solution for the after-sales system of lidar, including warranty period, loss determination, maintenance costs and so on.
This also means that at this stage, consumers pay for lidar, in exchange for only burying hardware conditions in advance for technological upgrades, as well as facilitating data collection and algorithm iteration of car companies.
If 2022 is the first year of mass production of vehicle-mounted lidar, then the term mass production refers to car companies, not consumers.