Article | Liu Junhong, Li Xiang, and Li Bin teamed up with "new top players in the car industry" Lei Jun to carry each other's sedan chairs, and the man in red Zhou Hongyi commented on various new car products with high emotional intelligence... In order to attract traffic, car c

Text｜Liu Junhong

Li Xiang and Li Bin teamed up with the "new top players in the car industry" Lei Jun to carry each other's sedan chairs, and Zhou Hongyi, the uncle in red, commented on various new car products with high emotional intelligence...

In order to gain traffic, this year the car company bosses competed Try to be "open" at all times.

But compared to the business of internet celebrity bosses, a secret war around intelligence has begun long before the auto show.

"You can use it wherever there is a road, and you can drive it wherever you go."

Prior to this, NIO began to push the global navigation assistance nop + urban area function on April 30. At 19:30 that night, NIO founder Li Bin also tested his own live broadcast, using the global navigation assistance during the evening rush hour in urban Beijing. After nearly two hours of driving, only 1.5 takeovers were achieved.

"Huawei Smart Driving is beyond my imagination. I took the to explore the world with m9 on the Fourth Ring Road (Beijing) for the past two days. I went on the main road and off the auxiliary road. I basically finished the journey." Zhou Hongyi, a new Internet celebrity in the car circle, compared the luxury brand After buying Cullinan , I am full of praise for domestically produced smart cars.

Huawei Smart Car Solutions Bu CEO Jin Yuzhi predicted, "2024 will be the first year of large-scale commercialization of smart driving."

"High-end smart driving" is accelerating to be installed in every smart car.

Taking 2024 as the watershed, car companies may no longer be divided into traditional and new forces. There will only be those that keep up with the wave of intelligence and those that cannot.

2024, high-end smart driving becomes standard equipment

"The turning point of automobile intelligence has arrived."

At the Qiankun launch conference of Huawei's new smart car solution brand, Jin Yuzhi said excitedly, "Among them, intelligence has become a key factor for users to buy cars. . Taking Wenjie as an example, more than 70% of users choose the high-end smart driving version.”

Now that smart driving has reached the stage of popularization, it cannot just be a high-end car configuration.

At the press conference of the high-end intelligent driving capability chip (Journey 6) the day before the auto show, facing the main model that accounts for 50% of China’s car sales, Horizon CEO Yu Kai said, “Consumers in the price range of 100,000 to 200,000 , I also want high-speed NOA and urban smart driving.”

With the assistance of technical service providers such as Huawei and Horizon, we saw at this auto show that almost all independent brand giants such as Great Wall, Changan , BYD , SAIC, GAC , FAW, BAIC , etc. They will step up efforts to implement high-end smart driving, and almost all smart cars will be equipped with high-end smart driving functions by default (or will soon be).

It seems that cars that are not yet available for high-end smart driving in 2024 have become "old antiques" of the previous era.

The reason why car companies attach so much importance to the implementation of high-end smart driving is that on the one hand, high-end smart driving (mainly refers to urban NOA functions) can better highlight the value of car intelligence. For example, in terms of sales, the purchase proportion of smart driving versions of Huawei Smart Car and Xpeng is higher than that of non-smart driving versions.

On the other hand, at the level of consumer experience, high-end smart driving covers a wider range of urban scenarios for daily driving. Consumers can easily confirm the experience through live demonstrations of car companies, "smart driving short videos" and actual test driving experiences. The degree of intelligence of Taiwan’s cars. If a smart car's high-end smart driving performance is poor, it will be easily "labeled" by consumers as not smart enough.

In fact, smart cars have formed the current trend of "evaluating intelligence by smart driving", which has long been foreseen by players in the entire automotive industry. However, in the process of jointly pursuing intelligent driving capabilities, only a few players such as Huawei, Xiaopeng, BYD, Great Wall, Geely have taken the lead. If other players want to pursue the implementation of high-end smart driving in 2024, they still need smart driving suppliers to "help".

It is precisely because of the urgent need of car companies to implement high-end smart driving that smart driving suppliers at the Beijing Auto Show have also swept away the haze of the "intelligent driving winter" a few years ago and began to talk about fixing and implementation.

Under the general trend of popularization of high-end smart driving, a new round of involution is also unfolding.

Among them, Huawei and SenseTime are benchmarking against and Tesla , taking the lead in the "end-to-end" technology of smart driving.

has updated the bev+god (similar to occupying the network) solution of the previous generation ads 2.0 to the god/pdp network solution of ads 3.0 on the smart car solution brand (Qiankun) just released by Huawei Auto.

Compared with the old solution, Huawei’s updates in ads 3.0 mainly focus on two aspects. First,

merges the two large models, bev, which were originally used to "see the world", and god, which understands obstacles, into one large model (god of ads 3.0) that can understand real driving scenarios. After the large model is "combined into one", Huawei Smart Driving can save more computing power.

The second is that on the pdp network, Huawei has upgraded the decision-making module that was originally "full of various detailed rules" into an end-to-end regulation, allowing smart driving to directly output to the car based on the real driving cases of "experienced drivers" Control module, thus realizing that smart driving considers integrity when operating in complex scenes, and the riding experience becomes smoother.

Although Huawei’s smart driving solution this time has not yet achieved complete end-to-end (perception direct output control) integration. However, Huawei's split solution for perception and decision-making can locate the problem immediately when smart driving makes an error, thereby achieving efficient iteration.

Next door to the Huawei Auto BU booth, SenseTime not only expanded from just smart cabins to smart driving last year, but also believed that the smart driving solution should be completely end-to-end and "one step in place".

"The two modules of perception and decision-making build a large model framework respectively, which is easier to implement. However, for the perception and decision-making of the 'two-stage' end-to-end solution, the information transmitted between the two models is artificially defined explicit information , data transmission will still be filtered and lost. "

Thanks to its advantages in the field of large models, SenseTime integrated perception, decision-making, planning and other modules on the end-to-end autonomous driving model on display - uniad. Into the end-to-end model of full-stack transformer. Let the intelligent driving system directly output the control results in "what you see is what you get", thereby reducing human intervention in the operation of the intelligent driving model.

Moreover, in the subsequent iterations of the large intelligent driving model, by continuously "feeding data", the efficiency of manual code writing can also be increased, and then under the generalization and "emergence" characteristics of the large model, the intelligent driving system can Quickly achieve “opening nationwide”.

Compared with the high-end smart driving solutions that are software-oriented and provided by Huawei and SenseTime, Horizon and Black Sesame provide car companies with different implementation ideas from the perspective of "all-inclusive software and hardware". As Jiyue Automotive CFO Liu Jining said during the auto show, "The second half of automobiles is about intelligence. Specifically, by 2024, intelligence will be high-end intelligent driving, which must be available out of the box."

At the booth, unlike the noisy booths of surrounding car companies, Horizon displayed a total of 3 display cabinets of smart driving controllers that are already in the mass production stage, focusing on "low-key luxury."

Horizon’s cooperation with car companies is based on software algorithm adaptation based on hardware (mainly chips and domain controllers). The Journey 6 series smart driving chips released by Horizon during the auto show are divided into different levels according to the computing power from 10+tops to 560tops according to the smart driving capabilities that car companies need to implement.

Among them, in response to the demand for high-end intelligent driving (mainly urban noa), Horizon has equipped the Journey 6p with more than two pieces of NVIDIA orin-x computing power, leaving space for car companies in the implementation and iteration of high-end intelligent driving. A rather "rich" space.

Moreover, Horizon has also built a complete dynamic, static, and OCC (occupied network) perception architecture in the all-scenario intelligent driving solution superdrive released this time, providing car companies with a "packaged" solution of chip + algorithm.

In addition, in order to "take care of" the different "hands-on capabilities" of car companies, Horizon has also cooperated with Bosch, Luxshare and other automotive tier 1 companies to create complete smart driving hardware solutions, allowing car companies to achieve "out-of-the-box use".

In the end, Horizon has built the "widest circle of friends" among Chinese smart driving manufacturers. It has now accumulated more than 230 designated models for cooperation, and it can be seen from the curve that the implementation of smart driving will begin to accelerate in 2022, and it has embarked on a steep journey. acceleration curve.

On the other hand, compared to Horizon’s “one-stop smart driving” solution, Black Sesame is thinking about the possibility of implementing high-end smart driving and a single-chip platform to integrate cabin and driving.

At this auto show, Black Sesame brought two mass-produced chips of the Wudang series c1200. Among them, c1236 is mainly aimed at single-chip high-end intelligent driving solutions, which can realize single-chip support for NOA parking and parking integration. On another product, the c1296, Black Sesame has installed an additional MPU (microprocessor) module on the smart driving chip platform.

By operating in this way, although the c1296 cannot reach a high-end level in smart driving functions, car companies can simultaneously realize "light" smart driving and cockpit functions through one chip, thereby reducing the amount of car wiring harnesses, reducing car costs and increasing battery life. the goal of.

Facing the future, Black Sesame believes that the integration of "car brains" at the chip level is the ultimate route to future car intelligence. "Market demand drives application exploration, which will push the technology side to use less cost, lower resources, and smaller devices." cost to achieve application maturity," Ding Ding, vice president of Black Sesame Intelligent Products, said at the auto show. The advancement of

smart cars at the smart driving level also has the power of autonomous driving companies that have entered smart driving from L4.

Light Cone Intelligence was seen at the WeRide Zhixing booth. Among the unmanned vehicles on display, there were two Chery Xingtu smart cars. According to the staff, the Chery Xingtu Xingyuan es and et mass-produced vehicles on site are equipped with high-end driving solutions jointly developed by WeRide and Bosch. The wave of implementation of

high-end intelligent driving has made intelligent driving companies that were once unable to implement L4 feel the "warmth" of large-scale commercialization.

Judging from the overall situation of the Beijing Auto Show, although smart driving suppliers have different pursuits of technology, their plans to accelerate the implementation of high-end smart driving in 2024 are very consistent.

Moreover, based on the cost of solutions given by various suppliers at this exhibition, the price of high-end smart driving models in the 100,000 to 200,000 yuan range is only 3% to 5%.

This also means that with the maturity of high-end smart driving technology and the involution of smart prices, most smart cars will have the opportunity to be launched in high-end smart driving in 2024.

After it is expected that cars will soon complete the intelligent transformation of travel aspects, consumers have begun to pay more and more attention to the intelligent experience of life in the car.

Large model "gets on the car", and the interaction of cockpit intelligence is smarter.

Large model is the trend of landing smart cockpit. Car companies around the world have already reached a consensus.

Overseas, at CES 2024, Volkswagen, Mercedes-Benz , BMW have used large models to create cockpit personal assistants. In China, they are the Ideal Mind GPT, NIO’s Nomi GPT, Volcano Engine’s large cockpit model and Tencent Smart Travel’s large automotive industry model, etc. Wang Xiaogang, chief scientist of SenseTime Technology, believes that the reason why

large models are "onboard" is that "after the large models are 'onboard', they can combine the single-point AI functions of the cockpit and automatically call the cockpit software and hardware resources. Multi-modal large models It can realize multi-sensory interaction between people and vehicles, significantly improving the smoothness and naturalness of interaction."

At this Beijing Auto Show, suppliers of automotive smart cockpits gave some of their latest thoughts on the "getting on the car" of large models.

Among them, SenseTime gave the "most sci-fi" answer of a large model landing cockpit in the smart cockpit using eye and gesture interaction similar to Apple's Vision Pro.

SenseTime's ability to interact with eyes comes from the accumulation of DMS (driver monitoring system, which can identify fatigue driving) in the car. With the improvement of recognition accuracy, SenseTime has achieved precise eye interaction with the driver's gaze on the central control screen, or the rear row users looking at the rear screen.

's "look and point" interaction method is very ideal, allowing the driver to control the car while driving without reaching out to touch the screen of the car, which improves driving safety. However, regarding the accuracy of the specific operation, Light Cone Intelligence will share it with everyone after experiencing and evaluating it.

On the other side, in terms of 3D gesture interaction, SenseTime has also designed an end-to-end recognition network for gesture recognition based on a large amount of real-person data. Users can perform car and machine operations that previously required touch "over the air". In this way, users in the back can more easily control the screen in front, control music playback, switch maps, etc.

From the new interactive technology, we can see that cars create a more three-dimensional operating space for the implementation of AI. Interactivity draws a lot of essence from PCs and mobile phones, and also creates differentiation in practicality.

In addition to interaction, the landing of the large model of the cockpit is also a highlight of this year's large models.

Tencent Smart Travel relied on the Hunyuan large model this year and upgraded the new large cockpit model, which can do intelligent questions and answers on car AI manuals, such as professional car use/car maintenance questions and answers for this model, and proactively provide vehicle condition reminders, maintenance suggestions, etc.; based on With the app agent capability, it only takes 1 week to learn and skillfully use 1,000+ in-vehicle applications, realize in-vehicle app/app in-depth voice interaction, and help users complete various in-vehicle service operations, such as automatically ordering coffee, booking movie tickets, Make reservations for valet services and other operations.

In the Qiankun smart car solution, Huawei Automobile has trained its own large model of the automotive industry - Qianwu engine based on the Pangu large model. With the blessing of this AI engine, the Hongmeng cockpit is in the car engine , audio, display and other links have been iteratively upgraded again.

At the car-machine level, based on Huawei’s self-developed Hongmeng system, Huawei’s car-machine can not only realize smooth connection of multiple devices, but also support native loading of most commonly used apps. Compared with other car companies that have “only a few” apps on their cars, and everything you want to watch on your cars, you have to cast your phone to the screen. Hongmeng’s cars are “way ahead” in terms of application ecology.

In terms of sound and display functions, Huawei also uses AI to adaptively process videos, automatically switching video sounds into 3D sound effects. Users can follow the picture and truly hear sounds from all sides of the picture. In addition, Huawei also uses AI to perform "directional noise reduction" in the cockpit space, so that listening to music in the front row and playing games in the back row will not disturb each other.

In addition to Tencent, Huawei and SenseTime, vehicle manufacturers are also experimenting with more possibilities in the interaction and experience of smart cockpits at this auto show.

For example, Great Wall Motors has also deeply applied AI large models in the coffee os 3 smart space released this time. Great Wall Motors' smart cockpit also implements interactive technology similar to SenseTime gesture control. Users can complete a series of operations such as playback, switching, and selection through specific actions such as pinching, dragging, and fisting. In addition, Great Wall Motors has also equipped the and Haval series models with the Coffee Life app. The car can connect to smart peripherals such as smart water cups, child seats, and car refrigerators to realize functions such as warming milk, making tea, and controlling temperature.

In addition, recommendations based on geographical location are also a new highlight. Zhiji Auto cooperated with Volcano Engine in the field of large-scale model application and algorithm to create city drive, which can filter and push local life content such as special food, popular attractions, cultural landmarks, etc. in real time according to the user's personalized preferences, making travel from "me" "Go" expands to "Take me there".

Overall, automobile smart cockpits and traditional car-machine function-based interaction models have become farther and farther away, and they are far ahead in terms of intelligence and richness of functions and applications.

Fang Yunzhou, founder of Nezha Automobile , believes that "the big model 'getting on the car' has now brought changes from 0 to 1, and will achieve greater changes from 1 to 100 in the future. People will hand over driving to cars. Smart spaces take care of other things”.

It is foreseeable that car companies will continue to increase their investment in automotive intelligence in order to stay ahead of their opponents at every stage, and a fully intelligent "rally" is about to begin.

Intelligent warfare beyond the entire vehicle

We can see that the "supply line" required to build a smart car is being "stretched" longer and longer.

In order to maintain long-term intelligent competitiveness, the cost expenditures of car companies are rising.The gap between the outstanding and the underachievers is also "expanding" at a speed visible to the naked eye.

"It even feels like the century-old advantages of joint venture brands have collapsed in the era of smart cars." A senior automobile industry analyst lamented about Guangcone Intelligence, "Even the new car-making forces that once had great sales have not increased investment in intelligence in the past two years, and they immediately felt lonely at auto shows."

On the one hand, it continues to "go shopping" in the wave of price cuts, and on the other hand, it must maintain continuous expenditure on intelligence. Even if car companies know that they have to spend this money, they will always look at the market for solutions that are more cost-effective than building their own intelligent computing centers and spending billions each year.

Therefore, the capabilities and base of large AI model manufacturers (mainly cloud manufacturers) are also the object of collective attention of car companies.

The combination of automotive intelligence and ai cloud, with the implementation and continuous iteration of high-end intelligent driving, the person in charge of intelligent driving who comes from a car company knows better how to mobilize this demand and anxiety.

NVIDIA Global Vice President Wu Xinzhou said, "After entering the era of AI-defined automobiles, the development model of intelligent automobile driving has changed, forming a closed loop of cloud + car development - intelligent driving models are trained and processed in the cloud with large computing power. Verification, and then deployment to the vehicle end, completes application and data collection and backhaul, and promotes model training in the cloud."

For example, as mentioned above, Huawei's smart car solution Qiankun 3.0, learns 30 million kilometers every day, and the model can be iterated every 5 days. The speed of is really scary.

In addition to intelligent driving training, car companies' research and development, marketing, office work, etc. all need large models for digital intelligence acceleration.

For example, Tencent's large model of the automobile industry can assist software engineers in writing code in the research and development process; in the marketing process, it can assist intelligent customer service and sales AI assistants; it can even build an AI engineering platform that car companies can directly use to develop their own AI applications required.

In the final analysis, the reason why car companies continue to increase investment is that smart cars are still far away from the "ultimate product". Perhaps within a few years, car companies will be in the midst of price cuts and increased investments in intelligence. But what is gratifying is that Chinese automobile brands will also build insurmountable barriers in their pursuit of intelligent technology.