Autopilot Route Dispute: The Auto Industry Also Has The Law Of True Flavor

In the field of automatic driving, Musk's obsession has always been well known. Musk's personal favorite pure vision solutions, the laser radar disdain, the reason is also very straightforward: from the "first principle", the best automatic driving should learn from the best human drivers, and drivers driving only rely on the eyes to discriminate, so the automatic driving should rely on vision, rather than additional sensors.
"Humans drive by sight, not by eyes firing lasers." Based on this theory, Tesla has been hoping to realize Autopilot with a pure vision solution, eschewing devices such as LIDAR.
The ideal is rich, the reality is bone-chilling. The pure vision route is not adopted by most car companies. At present, whether it is azure, small peng, ideal, weimar, wisdom, and other emerging enterprises, or the great wall, faw and other traditional car companies, have in the new models equipped with lidar, forming the camera, millimeter wave radar, lidar and other multi-sensor fusion system, rather than relying on the "what you see is what you get" of the depth of the visual learning.
Ideal Car even suggests that the role of LiDAR is similar to that of an airbag, although it can realize autonomous driving without it, but the visual learning scenarios always have limitations, and when encountering new scenarios, LiDAR can play a role in providing safety support.
The principle of the difference
Why is LiDAR, which is disdained by Musk, regarded as an "airbag"?
The principle of LiDAR is not complicated. It is summarized in one sentence: natural light in life is strong or weak, and lasers are the supplement to natural light in autonomous driving.
We fire a beam of laser light, which is reflected by the target and then collected by a receiving system. By measuring the running time of the reflected light, we can determine the distance to the target. Because lasers are highly monochromatic and collimated, the measurement results can directly reflect the object's distance orientation, depth information, reflectivity, etc., thus generating an accurate digital elevation model.
Autonomous Driving "Route" Controversy: Is There a "Law of True Fragrance" in the Automotive Industry?

Compared to visual learning technology, LiDAR is not affected by bad weather, and can be used day and night, in rain, snow, wind and frost, with little impact on overall performance.
It seems that the cost of the vision camera itself is much lower than that of LiDAR, however, the cost of building a learning system and a lot of road tests is unaffordable for small-scale car companies. This is because the visual learning mode urgently needs to solve the problem of "what is photographed", which requires sufficient big data support. A giant like Tesla can achieve good results in visual recognition through neural network training, self-developed FSD chip, cloud-based unsupervised training and original "shadow mode" analysis, but there is only one Tesla in the world.
In addition, the issue of safety should not be ignored. The safety of Autopilot is a matter of corporate reputation, and an accident is enough to have a devastating impact on corporate public relations. Under such a premise, most companies have the mentality of "seeking stability". Even if the camera + millimeter wave + ultrasonic radar can solve 99% of the problems, most of the car companies, like Ideal Auto, in the visual recognition model to establish at the same time choose LiDAR as an invisible "airbag".
Market choice
Visual learning technical support related people, will naturally claim that "pure visual route is the future trend, is the embodiment of the return to the origin of the automatic driving"; while the production of LiDAR enterprise, will also publicize "LiDAR is essential to make up for the limitations of the analog visual ". There is no end in sight to this unprecedented debate about the future direction of autonomous driving.
In terms of technology, different companies are in different situations, and it's hard to say which solution is better. In fact, rather than saying that the pure vision route and the LiDAR route is a technological battle of superiority or inferiority, it is more like the reality of the supply and demand relationship, the fight for market share.
Then we might as well see how the market chooses:
SAIC's "explosive" ZhiJi LS6 model launched in October 2023, LIDAR sensor + NVIDIA Orin ZhiDao chip will appear as standard equipment; 2022 Volvo XC90 and Lumina cooperation, equipped with roof-mounted LIDAR technology; in 2022, Great Wall Moca DHT -PHEV LiDAR version also successfully launched; 2022 Azera ET7 equipped with LiDAR, jointly developed by Azera and Tudatron ......
Even Tesla, the toughest "bone" in the field of visual recognition, is suspected to have "self-doubt" about the visual route:
Recently, the governor of Mexico's Nuevo Leon state, Samuel Garcia, said during a trip to Shanghai that two Chinese suppliers of Tesla will invest nearly 1 billion dollars in Nuevo Leon state in northern Mexico, where Tesla is planning to build a new factory. And one of them is none other than LIDAR company Hosai Group.
Tesla's invitation to Hershey's has been a very interesting one, and Hershey's has "not yet finalized" its decision to accept the invitation to build a new factory in Mexico. Of course, some people believe that Tesla's move is not to set up LiDAR for cars, but to supply humanoid robots.
As far as the market is concerned, the average consumer is most concerned, not whether a car is pure vision mode, or LiDAR mode. Rather, the car itself as a "black box" integrated after the realization of the performance of autonomous driving, safety in the end how. Safe or unsafe driving, the market to buy or not to buy, is the hard indicator of the quality of the car.
When the industry outlook is shrouded in fog, most of the car companies, even if they pay extra costs, but also to be equipped with "true" LiDAR, visible LiDAR is still a good way to realize a higher level of automated driving.