Recently, Toyota Motor's subsidiary Woven Planet Holdings announced that it is developing autonomous driving technology that does not use expensive sensors such as lidar. In response to this disclosure, the outside world generally believes that toyota has joined the camp of pure visual autonomous driving technology.
As the industry leader of the autopilot lineup, Toyota's unexpected turn has caused the industry to be stunned: at the moment when the lidar "gets on the car" tide, the "betrayal" of the "old fox" Toyota undoubtedly caught a large number of followers off guard. Why did Toyota jump out of the "lidar camp" proposed by itself, and how should the debate over the route of autonomous driving technology be determined in the field of mass production and commercialization?
LiDAR VS pure visual perception, who is the final answer to autonomous driving?
As we all know, in the two major development routes of intelligent driving, there have long been two major routes of "lidar and pure visual perception":
Pure visual perception route selection with machine vision as the core, the use of millimeter wave radar + camera solutions to achieve automatic driving, mainly representing the United States Tesla's FSD, China's Baidu's Apollo Lite, Israel's Mobileye's SuperVision three mainstream pure visual perception programs; The lidar solution chooses to use high-precision map + lidar as the core sensor to achieve automatic driving, representing the company's Waymo, Baidu Apollo, Xiaoma Zhixing and so on. It can be said that except for Baidu Apollo, which chooses two routes in parallel, most other companies have chosen to bet on a single route.
The dispute between the two routes is due to the obvious differences in their advantages and disadvantages.
The lidar scheme can help the vehicle to achieve rapid 3D modeling under the existing technical conditions, accurately restore the road condition information, and form a program that the computer can quickly identify, process and respond quickly. However, due to the high cost of the lidar solution and the large demand for chip computing power, it is impossible to complete the installation on all levels of models at this stage, which makes Musk, who intends to make Tesla a "Volkswagen and Ford in a smart car", cannot stand it, for which he publicly stated that lidar is a solution that "only fools will use".
The advantages of the pure visual perception route are more obvious, which is not only lower cost, but also in line with the logic of the human eye, and will produce qualitative changes after the data accumulation reaches a certain scale, until it reaches the automatic driving performance of the shoulder-to-shoulder and beyond the lidar scheme, which is a more technologically advanced solution. To this end, Wu Xinzhou, vice president of Xiaopeng Automobile, said bluntly: "In the long run, vision is omnipotent."
In fact, the potential of pure visual solutions has become a reality in practice: as the only pure visual L4 level urban road automatic driving solution in China, Baidu's Apollo Lite has the ability to achieve urban L4 level automatic driving only by pure visual perception, and in the field of mass production, it has also helped a series of OEMs to achieve dimensionality reduction commercialization of L4 capabilities on autonomous parking products AVP and pilot assistance driving products ANP.
However, the pure vision scheme is not perfect, and in harsh environments, the difficulty of the camera to complete the perceptual task will also increase. If it is applied to business at this stage, car companies that are more responsible for consumers must give them hardware redundancy.
In short, no matter what the route, in the field of mass production, we should focus on safety. Under this wave of debate over the new round of autonomous driving routes caused by Toyota, in the face of endless views, consumers cannot distinguish and do not need to distinguish, the market needs only a smarter and safer intelligent car, and the route of automatic driving should not be "in line".
So, what is the answer that the market needs?
Why is it that the "pure visual route and equipped with lidar" scheme may be the most worthwhile route to try at the current mass production?
Although there is a debate among major manufacturers about the route of autonomous driving landing, the problem to be solved is the same: in the end, it is necessary to answer the problem of how to mass-produce high-level autonomous driving.
This road is like the climb of Mount Everest, whether you choose the south slope or the north slope to reach the top, you will eventually come to the only way to the top: "Hillary Steps". The battle of routes is not between you and me, but to learn from each other's strengths, and they are all tools to achieve high-end automatic driving mass production.
That is to say, although the pure visual scheme is more advanced, it is not an exclusive relationship with the lidar scheme, nor is it a pure subordinate or complementary relationship, from the perspective of mass production and commercial use, the pure visual scheme must have redundancy to supplement the blindness.
So, without considering the cost, only considering the safety, the smart car is full of lidar, is everything going to be fine? After all, almost all flagship smart car manufacturers released around 2022 have been equipped with lidar as standard, and even models with 4 lidar have appeared.
Still not.
In response to the above question, Wang Liang, chairman of Baidu Apollo Technical Committee, has publicly stated: "In the traditional strategy of lidar as the mainstay and vision as the supplement, the problems and defects of visual perception itself are not fully exposed under the cover of radar perception. That is to say, whether it is a lidar route or a pure visual perception route, every autonomous vehicle cannot avoid solving the problem of visual perception.
Zhang Yaqin, Chair Professor of Intelligent Science and Dean of the Institute of Intelligent Industry of Tsinghua University, academician of the Chinese Academy of Engineering, and academician of the American Academy of Arts and Sciences, bluntly said at the 2021 Artificial Intelligence Computing Conference (AICC 2021) that the combination of vision and lidar is the best way to (land on automatic driving).
From industry practice to academic research, they all affirm the route of "vision-based, lidar provides safety redundancy", and Jidu Automobile has the advantages of this fusion technology.
As a new car manufacturer that directly inherits Baidu Apollo's leading intelligent driving technology and Robotaxi's 25 million kilometers 0 accident road test data accumulation advantages, Jidu can not only use the nutrients of "Apollo Lite", China's only pure visual L4-level urban road automatic driving solution, but also has the innate conditions of "starting point is the end of others" on the pure visual perception route, and also has a very deep reserve of lidar technology solutions to provide redundancy for the blindness of vehicles. After all, as mentioned above, Baidu is one of the few self-driving companies that walk on two legs with a "lidar route + pure visual perception route". This is rare in the world.
When the pure visual route of the bully better understands the value of "route integration", the first automotive robot equipped with lidar will also become a benchmark for autonomous driving mass production products.
In fact, just in January this year, Jidu has integrated high-speed, urban dual-domain multi-scenario autonomous driving functions on its SIMUCar (software-integrated simulation prototype), which has verified the safety and stability of high-level automatic driving capabilities in mass production in advance. Although the route chosen by Jidu has not yet been fully disclosed, it is speculated through some of its early vehicle maps that the probability of using "pure visual perception to achieve automatic driving and lidar redundancy to fill the blindness" scheme is not small. Once this speculation is confirmed, Jidu's first automotive robot is bound to become a benchmark for autonomous driving mass production products.
postscript
In Alice in Wonderland, the Queen of The Peach once said, "In our country, at your current speed, you can only stay where you are; If you're going to get to another place, you'll have to run at twice as fast as you are now. ”
In this "bayonet red" autonomous driving mass production competition, whoever has a deeper and more comprehensive technical reserve can go further. It can be said that although this round of competition has just begun, the set of autonomous driving "seed players" has not only obtained a ticket to the final, but is more likely to provide the industry with new ideas and new paradigms for autonomous driving mass production:
Pure vision technology solutions dominate autonomous driving, and lidar provides safety redundancy. The core question may not be how many lidars are carried, but who can start from the actual needs and experiences of users, so that automatic driving can truly become a technology that everyone can enjoy, rather than blindly "piling up" for users.