Every software and hardware technology change is reconstructing intelligence.
In July 2016, the Internet car Roewe RX5 jointly developed by SAIC Motor and Alibaba was officially launched. This is regarded as the first large-scale introduction of the Internet ecology in the Chinese passenger car market and the first upgrade of traditional fuel vehicles at the software level. A few months later, Intel acquired Mobileye for US$15.3 billion, and ADAS solutions based on visual perception became the leading players in the market.
This is regarded as the 1.0 era of smart cars, and the typical feature is the upgrade of functions and user experience on the basis of traditional automotive electronic architecture. From distributed ECU to domain controller architecture (especially the substantial increase in chip computing power), it has become a symbol of the 2.0 era of smart cars.
Whether it is Tesla's FSD or Xpeng's first mass-produced Nvidia Xavier, the upgrade of intelligent driving functions has entered a new cycle. At the same time, Qualcomm's third-generation Snapdragon automotive digital cockpit platform fully drives the hardware architecture upgrade of traditional smart cockpits.
In the 3.0 era of smart cars, cross-domain integration has become a label. Central computing + regional control, the further integration of smart driving and cockpit, and the further upgrade of the computing platform are the main features. And another larger cross-domain track is also emerging.
This week, BYD officially released the cloud car technology , aiming intelligence at body control (including intelligent damping body control system, intelligent air body control system, and intelligent hydraulic body control system). Among them, cloud car-X can use binocular cameras or The lidar is used to identify the ups and downs of the road ahead, and the state of the suspension system is adjusted proactively in a predictive manner, which is commonly referred to as the magic carpet suspension.
These technologies are not the first in the industry. In fact, from air suspension to magic carpet suspension, they have been mass-produced in recent years. For example, both Ideal L9 and Ideal L8 are equipped with magic carpet air suspension as standard, which can realize 60,000 perceptions and 6,000 active adjustments per minute by fusing 26 sensor signals in the car.
The deep integration of "smart driving collaboration" body control and ADAS is also regarded as another main line of the smart car 3.0 era. For example, the perception system of cloud car technology includes 57 perception units in 5 dimensions for perception; it has cloud data perception, road surface perception, driving behavior perception, body posture perception, system status perception, etc.
Behind this technology is binocular stereo vision, which has been regarded as a niche technology route in the past.
From the perspective of Yuan Oak Technology , the binocular camera scans the road ahead in advance to obtain images and 3D point clouds, combined with the MO-NET neural network and DeepFusion technology architecture independently created by Yuan Oak, the flatness and unevenness of the road in front of the vehicle and foreign objects Conduct detection, output highly reliable and high-precision azimuth, size and elevation information, and output relevant structured result data to the active suspension control system of the vehicle chassis, so as to realize the function of shock absorption and automatic adjustment of driving mode, so as to comprehensively improve the driving experience of the vehicle Comfort and chassis safety.
Prior to this, Yuan Oak Technology has received a number of leading self-owned brand car companies' fixed-point notices. The two parties will cooperate to develop road preview system (magic carpet) products, and mass production will start in 2023.
At the same time, this system can be reused in the mainstream driving intelligent driving functions, and extended to the integration of driving and parking. At the same time, the height limit and width limit can be collected through the area detection function and high-precision map drawing, which is also the strength of binocular stereo vision.
This is the only way out for current automakers to popularize and upgrade intelligence while controlling costs. Technology (software and hardware) reuse is the only way out. "Solving customer pain points in a cost-effective way" will also be the vane of a new round of intelligent competition cycle.
At present, in the high-end intelligent driving track, some manufacturers continue to strengthen pure visual perception capabilities, such as Tesla and Mobileye, while relying on data training iterations; the other part seeks to use binocular stereo vision + lidar + high-resolution Radar to achieve, for example, Lexus, Mercedes-Benz.
In addition, the magic carpet suspension function of the new generation of Mercedes-Benz S-Class is also based on the front-view binocular stereo vision solution, which provides road scanning function and coordinates control with the air suspension system. This is also the current mainstream trend in the automotive industry.
, duration 00:35
In the eyes of industry insiders, detection distance and accuracy are the key to higher-level automatic assisted driving. Among them, the detection distance, which has always been regarded as the "weakness" of binocular stereo vision, is also rapidly evolving and improving. "For long-distance detection, it can be based on software algorithm innovation to achieve a wide baseline configuration with a large distance between two cameras, while allowing mechanical tolerances and active compensation functions."
From the perspective of the overall perception scheme of binocular applications, compared with monocular, it does not increase the cost at the system development level; on the contrary, due to the output of dense point clouds (covering a range of 150-200m and high accuracy), it can With lidar and 4D millimeter-wave radar, it can achieve real original data-level fusion and achieve the effect of real perception redundancy.
"Vehicle-mounted applications have two characteristics: one is that the distance measurement distance is relatively long, and the other is high reliability requirements." According to Lu Yaojie, CEO of Yuan Oak Technology, from the beginning of the company's establishment, the company has improved the distance measurement distance from the algorithm itself capabilities, and at the same time solve the cost optimization of big data processing through self-developed chips.
In the view of Gaogong Intelligent Vehicle Research Institute, in the next few years, only the intelligent application upgrade of the chassis such as air suspension will have the opportunity to drive the binocular stereo vision technology to realize the real large-scale boarding. At the same time, based on the requirements of technology reuse and overall system cost reduction, the integrated solution of driving + chassis intelligent control is realized.
The data shows that in the Chinese market (excluding imports and exports) in 2022, 280,100 passenger cars will be equipped with air suspension as standard equipment, a year-on-year increase of 142.3%. Among them, the standard ADAS system accounted for more than 90%. From the perspective of price range distribution, the average price of new cars with pre-installed standard air suspension will drop from 575,200 yuan in 2021 to 553,300 yuan in 2022, and some models have dropped to 300,000 yuan.
With the accelerated mass production of domestic suppliers in China, increased localization of foreign suppliers, and the bright support of air suspension for new car marketing, it is expected that the annual delivery scale of passenger car air suspension systems in the Chinese market is expected to sprint to 3 million units in 2025.
And just before BYD released the cloud car technology this time, the company has announced that a number of subsequent models will be equipped with a front-view binocular stereo vision solution. Considering the current market share of BYD in the new energy vehicle track and the characteristics of high intelligent configuration, it will undoubtedly drive a new wave of "binocular stereo vision + magic carpet suspension".
This means that from the perspective of the system rather than a single sensor, the advantage of binocular stereo vision is that it can reduce the difficulty and cost of collaborative data fusion development of other sensors and perception systems. "This will open up entirely new perceptual combinatorial design options and reduce full life cycle development costs from a system perspective."