Toyota and Tesla chose the pure visual route, are they in the wrong camp?
In the field of autonomous driving technology, there has always been a dispute between computer vision and lidar.
Recently, Toyota subsidiary Woven Planet announced that the company will try to develop autonomous driving without using expensive sensors and only equipped with low-cost cameras. Tesla, which has always adhered to the visual route, has ushered in a new ally, and the pure visual route supported by the two giants of Toyota and Tesla can win a big victory in the perception route in the field of automatic driving?
For a long time, Tesla, Mobileye and other companies that adhere to the first principle have believed that the use of cameras + AI intelligent algorithms is enough, and Tesla CEO Musk has publicly stated that "fools use lidar."
The cost factor is clearly a major reason why Musk is attacking the lidar technology route. For example, in the early days, Velodyne's 128-line lidar sold for as much as $80,000, and the most basic 16-line lidar cost $4,000.
The Tesla automatic assistance standard suite Autopilot contains 8 cameras, 1 millimeter wave radar and 12 ultrasonic radars, which is obviously not considered for Tesla, which reduces the cost of the vehicle to challenge the fuel car, and the expensive lidar is not in the scope.
Toyota chooses the same route is also based on cost considerations, in the view of Woven Planet, the high cost of sensors such as lidar and millimeter-wave radar has greatly limited Toyota's road to automatic driving research and development, and the relatively low-cost pure visual route has become the best choice.
We know that "vision" is the source of perception of intelligent driving, similar to the role of the driver's "glasses" and "ears". It is undeniable that the perception ability of lidar is stronger than the camera, and the road to L3 is inseparable from the configuration of lidar, but with the technological progress and the scale effect brought about by mass production, the current price of vehicle-grade lidar has dropped to hundreds of dollars, which is already within the acceptable range of vehicle companies.
From the perspective of the domestic intelligent car market, lidar has become the consensus of most car companies on a large scale, in a sense, whether it is equipped with lidar or not has become a standard for measuring the performance of smart cars. Taking "Wei Xiaoli" as an example, Models such as Weilai ET5 and ET7, Xiaopeng P5 and G9 and Ideal L9 will be equipped with lidar, and more and more companies will follow up under the "praise" of many car companies. You know, lidar was originally proposed by NVIDIA and Toyota, and autonomous driving technology research and development companies such as Google, Volkswagen and Baidu in the mainland are all fans of this camp. As early as the 2017 demonstration of the self-driving prototype, Toyota arranged a total of 7 lidars from velodyne on the roof and front of the car, and Toyota's anti-water is really confusing.
Toyota seems to be "derailed", but in fact, it has another plan
Toyota so much fanfare into the embrace of pure vision, it is very likely to be affected by Musk, Tesla through the camera on the market more than 1 million cars on the camera collection data, so that the AI platform fully self-learning, the owner of the avatar "tool man" for Tesla's autopilot research and development to provide video data, through the neural network to build a real-world three-dimensional vector space, to achieve the car's behavior and path planning, and generate control signals to transmit to the actuator, Then, through the perfect data closed-loop system and simulation platform, the continuous iteration of autonomous driving capabilities is realized.
Now, Toyota is using the same approach. In 2021, Toyota's global sales reached 10.4955 million vehicles, and its penetration rate and popularity far exceeded Tesla's. Taking a purely visual route, the amount of data that rolls out is considerable. "Abandoning" high-end sensors instead of cameras is more convenient to deploy through existing models, and its cost is greatly reduced by 90%. At the same time, taking advantage of its advantages as a large car manufacturer and obtaining a large amount of effective data can enable its autonomous driving system to conduct sufficient deep learning and further drive future research and development.
Auto Network Review: Toyota and Tesla are confused about the appearance of pure visual routes
According to the forecast of the consulting firm Yole, the penetration rate of L3-level autonomous driving in the global passenger car market will reach about 6% in 2025, which means that 6 million new cars will be equipped with lidar every year. In the future, as costs fall further, lidar will become the same standard as airbags.
Choosing a pure visual route, the camera is more likely to make mistakes in the face of backlight, darkness, specular reflection and other situations that the human eye cannot recognize. The reason why Tesla frequently has autonomous driving safety accidents is mainly due to the lack of lidar, which is the necessary safety redundancy.
The lidar of the vehicle regulation level needs to undergo environmental tests, high and low temperatures, mechanical shock, vibration, salt spray and other tests before it can be put on the car before it can be loaded, which is self-evident. And the price is no longer an unacceptable category for car companies in the past.
Taking Sagitar Juchuang's second-generation intelligent solid-state lidar RS-LiDAR-M1 (M1 for short) as an example, the batch delivery price of M1 has dropped to between $200-500. Lucid Motors, "Wei Xiaoli", WM, Extreme Kr, Lotus and many other models have placed orders. Admittedly, the price problem is no longer a major pain point for the commercialization of lidar.
As for whether Tesla and Toyota will go all the way to black on the pure visual route?
In fact, although Musk has always scorned lidar, but behind the scenes is "really fragrant". In May last year, American netizens photographed a Tesla Model Y test vehicle equipped with lidar in Palm Beach, Florida, USA. Tesla has also signed a contract with lidar company Luminar as part of the agreement to sell tesla sensors for the Model Y.
Toyota, which was already a multi-sensor fan before the water reflux, set up a division called toyota Research Institute (TRI) in Silicon Valley in 2016, and a year later demonstrated a lexus-based LS600h test vehicle, which was equipped with up to 7 Velodyne lidars, in addition to six 16-line lidars on the front and roof.
It is true that Toyota and Tesla choose a pure visual route does not mean that they completely abandon the lidar route, in the context of the declining cost of lidar, perhaps in the future, Toyota, Tesla will once again renege on their promises, the future autonomous driving technology route is still on the road of multi-sensor fusion.