In the field of autonomous driving, in addition to the decrease in the use of lidar and millimeter-wave radar, the camera configuration has gradually shifted from multi-eye to binocular configuration, which is consistent with the mainstream development direction of the industry. According to the analysis of Huajin Securities, the camera, as a key visual sensor of the automotive autonomous driving system, is mainly responsible for collecting image information of the vehicle's surrounding environment, pedestrians, vehicles, etc. With the help of deep learning algorithms, the system is able to accurately identify the type, location, and movement of these objects.
Huajin Securities further pointed out that according to the number of on-board cameras, its scheme can be divided into three types: monocular, binocular and multi-eye. Among them, monocular cameras are the most common in the market, and their advantage is that the line of sight is long, but there is a limitation that they cannot obtain depth information. Binocular cameras are equipped with two cameras that can obtain depth information, but require more accurate camera calibration. On the basis of binocular cameras, multi-eye cameras can obtain more information, but they need more support in information fusion and computing, and the cost is relatively high. At present, most of the front-looking cameras of domestic mainstream manufacturers use multi-eye solutions. However, with the improvement of camera resolution and cost control considerations, the number of cameras equipped is gradually decreasing from more than two to two.
On-board computing power: gradually close to the actual demand
Huajin Securities pointed out that from the perspective of product release history, Huawei's intelligent driving computing platform has been upgraded from MDC 300F to MDC 810, and the CPU and computing power configuration has shown an upgrade trend. However, in practical applications, the on-board computing power has declined. For example, Huawei's ADS 1.0 has a computing power of 400 TOPS, while ADS 2.0 has a computing power of 200 TOPS. This suggests that although the computing power of the hardware is increasing, it may be adjusted according to the needs in practical applications.
The deployment of high-computing power hardware may be in preparation for upgrading to a higher level of autonomous driving in the future. However, at this stage, there may be a situation of redundancy of computing power. This not only introduces unnecessary power consumption, but can also affect the vehicle's range. For example, the NIO ET7 is equipped with 4 Orin chips, with a total computing power of 1016 TOPS, but in practice, only 2 Orin SoCs participate in real-time autonomous driving data processing. To a certain extent, this redundancy of computing power reflects a phenomenon in the current development of autonomous driving technology, that is, the computing power of hardware is ahead of the needs of practical applications.
Figure: Vehicle computing power is gradually approaching the actual demand (Source: Huajin Securities)
On-board computing power: from the standard configuration of the whole system to the "high + low" dual version configuration
Different consumers have different needs and usage scenarios for intelligent driving functions. Some consumers are pursuing high-end intelligent driving functions, such as urban NOA, which require high computing power to support complex perception, decision-making, and planning algorithms. On the other hand, some consumers pay more attention to basic driving assistance functions and have relatively low computing power requirements, and the low-computing power version is enough to meet their daily use. According to Huajin Securities, Huawei has explored the "Max+Pro" two solution modes on the Zhijie S7 released in November 2023, of which the Max series (including Max, Max+, Max RS, and Ultra versions) supports urban NCA and comes standard with one 192-line LiDAR, and the Pro version supports high-speed NCA and no LiDAR. In the field of intelligent driving, the competition among car companies is fierce. Offering a "high + low" dual version configuration can enrich the product line, target consumers with different budgets and needs, and increase market share.
From the perspective of the industry, since 2022, depending on whether it can support urban NOA, some mainstream intelligent driving manufacturers, including Xpeng and Auto Li, have begun to appear in the hardware configuration of the "Max + Pro" dual-version scheme, of which the Max version supports urban NOA and is equipped with higher computing power, which is ready for the future upgrade of the intelligent driving system, and the Pro version supports high-speed NOA and has a relatively low computing power configuration to ensure the realization of L2 intelligent driving, which is already more mature.
From the standard configuration of the whole series to the "high + low" dual version configuration, on the one hand, it enriches the product line of car companies, improves market competitiveness, can meet the needs of different consumers, and expand market share. On the other hand, it also puts forward higher requirements for the R&D and production capacity of car companies, and it is necessary to develop and optimize software algorithms on different computing platforms to ensure the performance and stability of products.
Related:
Huawei Intelligent Driving Solution Market Analysis Report (1)
Huawei Intelligent Driving Solution Market Analysis Report (2)
