High-end intelligent driving "drops" price "bomb"

As a representative of China's new local forces, Xpeng Motors' series of actions in "reducing costs and increasing efficiency" in intelligence this year may further widen the gap with other automakers, especially traditional joint venture brands. The latter originally planned to accelerate the upgrade of high-end intelligence through localized cooperation.

On November 7, the "world's first AI car" Xiaopeng P7+ was officially launched and sold with an official guide price starting from 186,800 yuan. The entire series is equipped with Qualcomm 8295P and dual Orin-X+ end-to-end large model for full-scene high-end intelligent driving. This is after the MONA M03 in August this year, Xiaopeng once again strengthened its new strategy of popularizing full-scene NOA for models below 200,000 yuan.

This is also the first time that Xiaopeng has launched a new model that no longer distinguishes between the Max and Pro versions. In the past, only the Max version was equipped with a laser radar, supporting the full-scenario intelligent driving functions of city navigation + high-speed NOA. The Pro version does not have a laser radar and only has high-speed NOA and basic assisted driving functions.

"We hope to achieve the same safety by removing the LiDAR when a person is driving." In He Xiaopeng's view, the end-to-end model makes this possible because, under human supervision, the upper limit of the visual solution is increased and the necessity of LiDAR is reduced.

In fact, the goal of reducing the cost of autonomous driving hardware and optimizing the BOM is one of the core strategies proposed by Xiaopeng Motors in 2023; "In the face of fierce industry competition and internal circulation, it is very important to have strong cost control capabilities." Xiaopeng Motors CEO He Xiaopeng admitted that cost control was not done well in the past.

In addition to platform modularization, intelligent driving is one of the core competitive advantages of Xiaopeng models and is also a key link in the cost reduction roadmap. From "computing power + lidar" to "computing power + end-to-end", the enhancement of visual perception capabilities is the decisive factor.

For example, the AI ​​Eagle Eye vision solution first used in the Xiaopeng P7+ is the first in the industry to adopt a single-pixel LOFIC architecture, which effectively improves the visual recognition capabilities in dark light, backlight and large light difference scenes. Public data shows that the perception distance of this vision solution has increased by 125% and the recognition speed has increased by 40%.

In addition, in order to make up for the shortcomings of traditional pure vision solutions (originally reinforced by lidar), Xiaopeng's self-developed cloud-based large model is trained with a large number of parameters (up to 80 times the number of vehicle-side parameters) to exhaust the long-tail problems in intelligent driving as much as possible to cover more driving scenarios.

At present, it is still unclear how much the overall system cost of the AI ​​Eagle Eye Vision solution can be reduced compared to the earlier LiDAR solution. Previously, He Xiaopeng publicly stated that the AI ​​Eagle Eye Vision solution and the LiDAR solution can achieve the same intelligent driving experience. In the future, the two solutions will be updated simultaneously to ensure the same experience.

This means that in the short term, the two solutions will continue to exist in parallel. Previously, RoboSense's innovative first-generation mid- and long-range LiDAR MX has also officially announced that it has been ordered for new cars from Xiaopeng Motors. Of course, it is a seamless upgrade of the previous generation of LiDAR solutions at a lower cost and with better performance.

At this point, Xiaopeng Motors has become the only automaker on the market that has both pure vision and lidar self-developed full-scenario NOA intelligent driving solutions in place. This has also left more room for the diversification of configurations of its subsequent models. At the same time, the price of high-end models (with additional lidar) also has a certain upward space, thereby increasing gross profit margins.

This also means that LiDAR may change from a "cost expenditure item" in the past to a new source of income and profit. In fact, before this, many car companies have already achieved the popularization of high-speed NOA by canceling LiDAR configuration. However, due to the existing capacity boundaries and safety redundancy considerations, these car companies have not aggressively launched pure visual urban NOA solutions.

In the view of Gaogong Intelligent Automobile Research Institute, Xiaopeng's new strategy on full-scenario NOA may soon be responded to and followed by the market. Especially for the market with a volume of less than 200,000 yuan, cost reduction is the only way to popularize high-end intelligent driving.

Data shows that currently, in the Chinese market, the price of high-speed NOA, which has relatively low hardware costs, has entered the 100,000 yuan range for standard models. In comparison, the price of urban NOA standard models has just entered the 200,000 yuan range.

"Automakers' full-stack self-research does provide more room for experimentation for the flexibility of vehicle solutions. In comparison, automakers that adopt third-party supplier solutions tend to be more conservative, and the trust in suppliers is also involved." In the view of some industry insiders, in the short term, the need to configure LiDAR will become a new "burden" for automakers that do not have the ability to develop high-end intelligent driving themselves, especially joint venture brand automakers.

In fact, as a few car companies such as Xiaopeng and Jiyue have taken the lead in removing lidar and fully betting on the end-to-end large-model technology route, those joint venture brand models that still have lidar on their heads and are preparing to debut in China's high-end intelligent driving market in 2025 may have "fallen behind" again.

So far, first-tier joint venture brands including Volkswagen, Toyota, GM, Ford, Mercedes-Benz, and Audi have finalized the lidar partners for new models in the Chinese market. Considering these manufacturers' conservative pricing of models and strict constraints on the bottom line of model profits, it may be difficult for them to compete with their Chinese counterparts.

In addition, as the implementation of Tesla's FSD pure vision solution in the Chinese market enters the final stage of the tough battle, it may once again trigger the "catfish effect" brought about by the localized production in China a few years ago.

For example, taking Model Y as an example, the price of the mid-priced model (long-range four-wheel drive version) plus the FSD optional package is about 350,000 yuan. This means that the market logic of equipping models below 300,000 yuan with LiDAR to provide full-scenario NOA is no longer valid.

In fact, this is not just one person’s opinion.

Previously, Richard Yu, Chairman of Huawei's Intelligent Automotive Solutions BU, said that Huawei's high-end intelligent driving models (equipped with laser radar) priced below 300,000 yuan are sold at a loss. This is why concepts such as no-map, light radar, and end-to-end have continuously impacted the intelligent driving track in the past year. The core appeal behind them is to "reduce costs and increase efficiency."

For example, the R&D architecture has been adjusted. The two departments of "perception" and "regulation and control" have been merged into an "end-to-end" system. According to industry insiders, the total amount of code will be significantly reduced from the classic architecture to the end-to-end architecture, and the core competitiveness of intelligent driving has also begun to shift from algorithms to data and computing power.

As a leader in intelligent driving, Li Liyun, vice president of Xiaopeng Motors, also publicly stated that the change in the intelligent driving competition paradigm means that "the way companies work has changed from piling up a lot of manpower, piling up scenarios, and writing rules to how to use a high-quality data flywheel system to solve problems."

Some car companies continue to increase configuration and increase sales.

This year, GAC Aion has continuously increased its investment in high-end intelligent driving. The second-generation Aion V and Aion RT models are both equipped with full-scenario NDA functions equipped with laser radar at the 150,000 yuan level. According to the company, "next year's intelligent driving costs will be reduced by half on the existing basis, and we will strive to make all Aion products equipped with high-end intelligent driving as standard within two to three years."

In the view of the company's top management, if they want to gain a foothold in the fiercely competitive A-class car market, "first, the product strength must be maximized, and second, the price competitiveness must be improved." Breaking the inherent perception of consumers that "smart driving models = high prices" has become the core strategy of this automaker to revive sales.

In addition, this year, with the support of Huawei Qiankun Intelligent Driving, Changan Deep Blue has also continued to increase its investment in the high-end intelligent driving market below 200,000 yuan. At the same time, another new energy sub-brand, Qiyuan, has also begun to deliver Changan's self-developed high-end intelligent driving, especially end-to-end large models, and it is expected that subsequent costs will be further optimized.

Monitoring data from Gaogong Intelligent Automobile Research Institute shows that from January to September this year, the delivery proportion of new energy passenger vehicles priced below 200,000 yuan in the Chinese market (accounting for the entire new energy vehicle market) reached 64.49%, an increase of about 3 percentage points year-on-year.