In the past few years, with the mass production and delivery of high-computing power automotive-grade SoCs and the trend of multi-sensor fusion solutions (to cope with high-end functions, especially urban NOA), intelligent driving solutions based on high-cost and high-performance "domain controller" architecture have become the focus of market attention.
According to the monitoring data of the Gaogong Intelligent Vehicle Research Institute, in 2023, the number of new cars delivered by the Chinese market (excluding import and export) passenger cars equipped with intelligent driving domain controllers as standard will exceed 2 million for the first time, a year-on-year increase of 63.86%. However, another set of data shows that the contribution of models with a price of more than 250,000 yuan accounts for more than 80%, highlighting the cost sinking pressure of the domain control architecture.
According to public information, at present, the cost of the L2 assisted driving front-view all-in-one machine solution (fusion of multiple sensors, which can support L2+ at most) has been as low as hundreds of yuan (pure vision), and in comparison, the domain control scheme is still at the level of thousands of yuan or even tens of thousands of yuan. For models below 250,000 yuan, the cost sensitivity is extremely high.
According to the data, taking China's passenger car market in 2023 as an example, the annual delivery volume of models below 250,000 yuan still accounts for more than 70% of the overall market, and even models below 150,000 yuan account for nearly 50% (in the range of 25-150,000 yuan, it is the game market of all-in-one machine solution, small and medium-sized computing power driving and parking integrated domain control and cabin driving integrated solution).
First, the forward-looking all-in-one machine, not to be underestimated
In recent years, with the gradual increase in regulatory requirements for new vehicle active safety (including L0-L2+, especially AEB) functions, the former emphasizes the "popularization" of policies and regulations, focusing on reducing accidents and fatalities, rather than improving the driving experience, unlike the market-driven characteristics of high-end ADAS such as NOA.
According to the monitoring data of the Gaogong Intelligent Vehicle Research Institute, in 2023, the proportion of passenger cars with less than 150,000 yuan in the Chinese market (excluding import and export) equipped with AEB as standard will only be 27.48%; In contrast, the standard configuration rate of models with more than 250,000 yuan has exceeded 90%. At the same time, the average delivery price of the standard equipment of Zhongda computing power domain control models is still as high as 363,300 yuan, which also leaves enough market space for the forward-looking all-in-one machine solution.
Entering 2024, the forward-looking all-in-one solution is still in an upward growth cycle; From January to June, the overall market increased by 11.75% year-on-year, of which the standard equipment of self-owned brand models increased by 15.99% year-on-year, continuing to outperform the growth rate of new car sales.
Another set of data shows that compared with the standard installation rate of ADAS (L0-L2) of joint venture (including wholly foreign-owned) brands, which is close to 80% (79.85%), independent brands are still lagging behind by nearly 30 percentage points. At the same time, the latter delivered more than 1.2 million new vehicles in the first half of this year.
This means that self-owned brand models will be the focus of market competition.
Among them, in the market of independent brand forward-looking all-in-one computing solution providers, thanks to the large-scale mass production of many independent brand head car companies such as BYD, Chery, Geely, and Changan in recent years, Horizon leads the way with a share of 33.73% in the first half of 2024, rising from the second place in the whole of last year to the top for the first time, and leading the second place by nearly 6 percentage points.
During this year's Beijing Auto Show, as the leader of the Chinese market, BYD Chairman Wang Chuanfu said bluntly that since the launch of the strategic cooperation with Horizon in March 2021, BYD has equipped millions of vehicles with the Horizon Journey family computing solution (covering two platforms: all-in-one machine and domain control).
Especially in the 8MP single-front video all-in-one market, Horizon is the first to provide the industry with high-performance and more cost-effective intelligent driving solutions. For example, the Horizon Matrix® Mono 3 monocular visual perception solution based on Journey 3 is the first to realize monocular visual matching and positioning, multi-source fusion positioning, and realize the decentralization of lane-level navigation and NOA functions.
According to the feedback from car companies, in addition to the advantages of combining software and hardware (with high-performance computing solutions and perception algorithms optimized for local scenarios), Horizon also has an open platform and a highly optimized tool chain, which supports customized development, and has the dual advantages of fast onboarding and flexible scalability.
At the same time, the upgrading of the stock market is also starting.
From the perspective of substitution and upgrading, less than 20% of the new AEB vehicles with standard AEB in the Chinese market in 2023 have completed the upgrade of the old and new solutions (8MP all-in-one machine or domain control architecture). Among them, the 8MP all-in-one machine can improve the pedestrian detection performance to about 100-150 meters; The 30-degree horizontal field of view can realize vehicle detection of about 500 meters and small target detection of about 180 meters.
For example, taking the 8 million pixel (based on Horizon Journey ®3) intelligent front-view camera all-in-one machine delivered by Continental Core Intelligent Driving delivered last year as an example, compared with the mainstream 2 million front-view camera in the market, on the basis of meeting the function of ADAS L2, it can well identify large curvature curves, identify other vehicles to grab the road earlier, and can meet the requirements of EU E-NCAP2023 and other regulations.
At the same time, the sensor extension based on 1VXR (a high-performance all-in-one computing solution, which can also be regarded as a mini version of domain control) can realize the NOA navigation assistance driving function (and support data backhaul). This means that compared with the NOA scheme of domain control architecture, the cost of getting on the car can be greatly reduced.
This means that the basic L2 can also continue to expand to L2+ and L2++ while most of the hardware remains unchanged. At the same time, it also solves the pain points of many car companies that are limited by cost constraints and cannot get on the high-end intelligent driving function.
Second, driven by laws and regulations, it is being implemented
"In the past, active safety and even basic L2 were almost disconnected from high-end intelligent driving," industry insiders said, behind this, not only the cost issue. On the one hand, some brands have strongly promoted high-end intelligence, resulting in the low-end intelligent driving market being left out in the cold; On the other hand, regulatory standards have not kept pace with technological upgrades.
Now, this situation is being broken.
Taking the European market as an example, from July 7, 2024, the new EU regulations on active safety functions of vehicles will come into force throughout the EU, covering all passenger cars and commercial vehicles. Among other things, the requirements for ISA Intelligent Speed Assist, AEB automatic emergency braking, DDAW driver fatigue and attention warning, and ELKS emergency lane keeping system functions are set out as standard.
In the Chinese market, so far, the Standardization Administration of the People's Republic of China and CNCAP have issued a series of test requirements and standards for the above functions, but they are not mandatory standards for the time being. The good news is that the mandatory national standard of "Technical Requirements and Test Methods for Automatic Emergency Braking System for Light Vehicles" is in the final stage of testing and verification.
This means that the era of mandatory standard equipment for basic ADAS functions represented by AEB automatic emergency braking is coming; At the same time, the new standard also greatly improves the performance requirements on the basis of the original test requirements, which indicates that the market will usher in new opportunities for both incremental and stock upgrades.
For example, taking the current recommended national standard GB/T 39901-2021 as an example, for stationary objects, the vehicle is required to drive at a speed of 30km/h without collision to be qualified; Moving targets require the vehicle to be qualified when driving at a speed of 50 km/h without a collision.
In fact, most of the mainstream AEB system designs in the Chinese market (in recent years, high-level NOAs based on large computing power domain control have generally greatly increased the upper limit of performance), are also developed based on the upper limit set by the above standards. For example, on NIO's first-generation intelligent driving platform, the relative speed between the vehicle and the vehicle in front is greater than 50 km/h, and if the vehicle in front is stationary or slow, there is a risk that it will not be able to brake.
At present, whether it is CNCAP, Euro NCAP or North American NHTSA, the AEB braking speed limit requirements for current vehicles are still increasing; For example, NHTSA made it clear this year that it plans to raise the AEB test standard (effective from September 2029) to require braking at speeds of up to 62 mph (about 100 km/h) to avoid collisions with vehicles in front.
3. Performance improvement, ongoing
This also brings the urgency of upgrading cost-effective all-in-one computing solutions.
For example, for the low-end intelligent driving market, the Journey 6B launched by Horizon is a forward-looking all-in-one machine solution that focuses on "extreme cost performance". At the same time, Sony is based on the world's first high-performance front-view perception scheme that supports up to 17 million pixels based on the Journey 6B, further breaking the performance constraints of 8 million pixels.
At the same time, with the computing power of the new Nash architecture BPU exceeding 10TOPS (the upper limit of the computing power (TOPS) of the world's first forward-looking all-in-one machine has been increased to double digits), as well as the built-in high-performance automotive-grade CPU providing more than 20K DMIPS computing power, and the 17 million pixel forward-looking perception solution, the maximum detection distance can reach 450 meters.
Behind the improvement of computing power, the direct benefit is the upgrade of the perception algorithm based on deep learning, which greatly improves the distance and accuracy of target and lane line recognition, and can effectively improve the recognition ability of special-shaped vehicles, two-wheeled vehicles, tunnel entrances and exits, bad weather and other different scenarios.
Product iteration also brings new market opportunities. As the first partners, Bosch and Denso Japan are also the leading players in the traditional all-in-one market. This means that Journey 6B has the opportunity to achieve a breakthrough from an independent brand to a joint venture brand in the all-in-one solution market.
Taking Denso as an example, in 2023, Toyota's front-view all-in-one machines from Toyota will be installed in the Chinese market alone, accounting for about 15% of the overall market share. Bosch has delivered more than 3.4 million units, which is enough to show the huge potential size of the alternative upgrade market.
At the same time, because Journey 6 has a unified hardware architecture, a unified tool chain and a unified software stack, car companies can achieve the fastest mass production efficiency from all-in-one machine to domain control, from active safety to full-scenario intelligent driving in the shortest possible time.
This also solves the problem that car companies are forced to choose two or more vendors with different architectures of SoCs (respectively for all-in-one and domain control) to achieve cost and function differences in the design of high and low ADAS functions. At the same time, the algorithm and software migration of the two SoCs is a significant hidden cost.
At the same time, with the further improvement of SoC performance, the traditional front-view all-in-one solution is also constantly optimized. For example, compared with the previous generation solution, the volume and system power consumption are greatly reduced, and the integrated structure of SiP modules and the expansion and upgrading of functions are supported, and OTA blessings are supported, which further reduces costs and increases efficiency.
The game between the all-in-one machine and the domain control solution is obviously also seeking a balance.
Interestingly, in order to enhance the safety redundancy of the high-end intelligent driving solution, we can see that since last year, many new models have undergone new changes in the scheme configuration. For example, the upgraded version of Volkswagen's new Tiguan L Pro, the IQ. The Pilot intelligent assisted driving system is equipped with two sets of solutions at the same time, the front-view all-in-one machine and the domain control system in the top-of-the-line version, which are safe and redundant with each other.
In the view of the Gaogong Intelligent Vehicle Research Institute, the high-performance forward-looking all-in-one machine + small and medium-sized computing power domain control has the opportunity to become the mainstream solution of the popular market of intelligent driving, to adapt to the needs of high and low configuration functions, and better meet the cost demands of car companies.