Related reading: 2023 Beijing Auto Show
"There are no competing car models, only smart cars that roll up to death".
Text: Liu Junhong
Li Xiang and Li Bin bound Lei Jun, the "new top class of the car rim", to carry each other's sedan chairs, and Zhou Hongyi, the uncle in red, commented on the new cars with high emotional intelligence......
For the sake of traffic, this year, the bigwigs of car companies are working harder than ever to "open".
But compared with the business of Internet celebrities, a dark war around intelligence has long been launched before the auto show.
"If you have a road, you can use it, and you can drive it anywhere. ”
On April 25, the first day of the 2024 Beijing Auto Show, Great Wall Motors debuted the Wei brand Blue Mountain Intelligent Driving Edition, continuing to compete for the right to speak for "high-end intelligent driving".
Prior to this, Weilai Automobile began to push the global pilot assistance NOP+ urban function on April 30, and at 19:30 in the evening of the same day, Li Bin, the founder of Weilai Automobile, also self-tested live broadcast, during the evening rush hour in Beijing's urban area, used the global pilot assisted driving for nearly two hours, and achieved only 1.5 takeovers.
"Huawei's intelligent driving is beyond my imagination, in the past two days, I have been sitting on the M9 in the fourth ring road (Beijing), on the main road and on the auxiliary road, and I have basically finished. Zhou Hongyi, a new Internet celebrity in the car circle, was full of praise for domestic intelligent driving after comparing the luxury brand Cullinan.
Jin Yuzhi, CEO of Huawei's Intelligent Vehicle Solution BU, predicted, "2024 will be the first year of large-scale commercial use of intelligent driving. ”
"High-end intelligent driving" is accelerating to be installed in every smart car.
With 2024 as the watershed, car companies may no longer distinguish between traditional and new forces, only those who keep up with the wave of intelligence, and those who can't.
In 2024, high-end intelligent driving will become the standard
"The inflection point of automotive intelligence has arrived. ”
At the launch of Qiankun, Huawei's new brand of intelligent vehicle solutions, Jin Yuzhi excitedly said, "Among them, intelligence has become a key factor for users to buy cars. For example, more than 70 percent of users choose the high-end intelligent driving version."
Jin Yuzhi, CEO of Huawei Intelligent Vehicle Solution BU
Now that intelligent driving has come to the stage of popularization, it can't just be the configuration of high-end cars.
In the high-end intelligent driving ability chip (Journey 6) press conference the day before the auto show, in the face of the main models that account for 50% of China's automobile sales, Yu Kai, CEO of Horizon, said, "Consumers in the price range of 100,000~200,000 also want high-speed NOA and urban intelligent driving."
With the assistance of technical service providers such as Huawei and Horizon Robotics, we saw at this auto show that almost all of the independent brand giants such as Great Wall, Changan, BYD, SAIC, GAC, FAW, AND BAIC WILL STEP UP THE IMPLEMENTATION OF HIGH-END INTELLIGENT DRIVING, AND ALMOST ALL SMART CARS ARE EQUIPPED WITH HIGH-END INTELLIGENT DRIVING FUNCTIONS BY DEFAULT (OR SOON).
It seems that the cars that cannot land high-end intelligent driving in 2024 have become "old antiques" of the previous era.
The reason why car companies attach so much importance to the landing of high-end intelligent driving is, on the one hand, that high-end intelligent driving (mainly referring to the urban NOA function) can highlight the value of intelligent vehicles. For example, in terms of sales, the proportion of Huawei's smart car and Xpeng's intelligent driving version is higher than that of the non-intelligent driving version.
On the other hand, at the level of consumer experience, high-end intelligent driving covers a wider range of urban scenarios for daily driving, and consumers can easily confirm the intelligence of a car from the live broadcast of car companies, "smart driving videos" and actual test drives. If the high-end intelligent driving performance of smart cars is not good, it will be easy for consumers to be "labeled" as not smart enough.
In fact, the formation of smart cars has long been foreseen by players in the entire automotive industry. However, in the process of jointly chasing intelligent driving capabilities, only a few players such as Huawei, Xiaopeng, BYD, Great Wall, and Geely have taken the lead. If other players want to chase the landing of high-end intelligent driving in 2024, they still need to "pull a hand" from intelligent driving suppliers.
It is also under the urgent demand of car companies to implement high-end intelligent driving, and the intelligent 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 fixed-point and landing.
Under the general trend of the popularization of high-end intelligent driving, a new round of involution is also unfolding.
Among them, Huawei and SenseTime are benchmarking against Tesla, leading the "end-to-end" technology of intelligent driving.
On the intelligent vehicle solution brand (Qiankun) just released by Huawei's car BU, the BEV+GOD (similar to occupancy network) solution of the previous generation ADS 2.0 has been updated to the GOD/PDP network solution of ADS 3.0.
Compared with the old solution, Huawei's ADS 3.0 update is mainly in two aspects.
The first is to merge the BEV, which was originally used to "see the world", and the GOD, which is used to understand obstacles, into a single large model (ADS 3.0 GOD) that can understand real driving scenarios. After the large model is "combined into one", Huawei's intelligent driving can save computing power even more.
Second, on the PDP network, Huawei has upgraded the decision-making module, which was originally "full of detailed rules", to end-to-end regulation, so that the intelligent driving can directly output to the vehicle control module according to the real driving cases of "old drivers", so that the operation of intelligent driving in complex scenarios is more holistic, and the riding experience becomes smoother.
Although Huawei's intelligent driving solution has not yet achieved complete end-to-end integration (perception, direct output control), it has not yet achieved complete end-to-end integration. However, Huawei's splitting solution for perception and decision-making can locate problems as soon as intelligent driving makes mistakes, enabling efficient iteration.
Next door to Huawei's car BU booth, SenseTime not only expanded from only doing smart cabins to smart driving last year, but also believes that the smart driving solution should be completely end-to-end "one-step".
"The two modules of perception and decision-making each build a large model framework, which is easier to implement. However, in the perception and decision-making of the 'two-stage' end-to-end scheme, the information transmitted between the two models is artificially defined explicit, and the data transmission will still be filtered and lost. ”
Benefiting from its advantages in the field of large models, SenseTime has integrated perception, decision-making, planning and other modules into the end-to-end model of the full-stack Transformer on UniAD, the end-to-end autonomous driving model on display. Let the intelligent driving system directly output the control results in "what you see is what you get", so as to reduce the intervention of human beings in the operation of the intelligent driving model.
In addition, in the subsequent iteration of the intelligent driving model, through continuous "feeding data", it can also increase the efficiency of manual code writing, and then under the generalization and "emergence" characteristics of the large model, the intelligent driving system can quickly realize "national opening".
SenseTime's end-to-end intelligent driving solution
Compared with the software-biased high-end intelligent driving solutions provided by Huawei and SenseTime, Horizon and Black Sesame provide different landing ideas for car companies from the perspective of "software and hardware all-inclusive". As Liu Jining, CFO of Jiyue Automobile, said during the auto show, "The second half of the car is intelligent, and by 2024, intelligence is high-end intelligent driving, which must be out of the box."
At the booth, different from the noise of the booths of surrounding car companies, Horizon displayed a full 3 showcases of intelligent driving controllers that have been in the mass production stage, focusing on a "low-key luxury".
Journey 6 series controller showcase in cooperation with major car brands and Tier1
Because the cooperation between Horizon and car companies is based on software algorithm adaptation based on hardware (mainly chips and domain controllers). On the Journey 6 series intelligent driving chips released by Horizon during the auto show, according to the intelligent driving capabilities that car companies need to land, they are divided into different levels according to the computing power from 10+ TOPS to 560TOPS.
Among them, in the demand for high-end intelligent driving (mainly referring to urban NOA), Horizon is equipped with more than two pieces of Nvidia Orin-X computing power on the journey 6P, leaving a considerable "rich" space for car companies in the landing and iteration of high-end intelligent driving.
In addition, Horizon has also built a complete dynamic, static, and OCC (occupancy network) perception architecture in SuperDrive, a full-scene intelligent driving solution released this time, providing a "packaged" solution of chip + algorithm for car companies.
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 Tier1 companies to create a complete intelligent driving hardware solution, so that car companies can achieve "out-of-the-box use".
In the end, Horizon has built the "widest circle of friends" of China's intelligent driving manufacturers, and has accumulated more than 230 cooperative fixed-point models, and it can be seen from the curve that the landing of intelligent driving will begin to accelerate from 2022, and it has come out of a steep acceleration curve.
Horizon high-end intelligent driving solution press conference, light cone intelligent shooting
On the other hand, compared with Horizon's "one-stop intelligent driving" solution, Black Sesame is thinking about the possibility of landing high-end intelligent driving and a single-chip platform to achieve the integration of cabin 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 the integration of NOA driving and parking with a single chip. On another product, C1296, Black Sesame has an additional MPU (microprocessor) module on the intelligent driving chip platform.
Although this operation makes C1296 unable to reach a high-end level in intelligent driving functions, car companies can realize the functions of "light" intelligent driving and cockpit at the same time through one chip, thereby reducing the amount of automotive wiring harness, and achieving the purpose of reducing vehicle costs and increasing battery life.
Black sesame booth at the Beijing Auto Show
Facing the future, Black Sesame believes that the integration of the "car brain" at the chip level is the ultimate route for future automotive intelligence, "Market demand will promote the application downturn, which will promote the technical side to achieve the maturity of the application with less cost, lower resources and less overhead", Ding Ding, vice president of Black Sesame Intelligent Products, said at the auto show.
The promotion of intelligent vehicles at the level of intelligent driving also has the power of autonomous driving companies that have cut into intelligent driving from L4.
Light Cone Intelligence saw at WeRide's booth that among the unmanned vehicles on display, there were two Chery Xingtu smart cars. According to the staff, the Chery Xingtuxing Era ES and ET production cars at the scene are equipped with high-end driving solutions cooperated by WeRide and Bosch.
The landing tide of high-end intelligent driving has made the intelligent driving companies that have been delayed due to L4 also feel the "warmth" of large-scale commercialization.
WeRide exhibited Chery Xingtu Interstellar Element ES and ET production vehicles
Judging from the overall situation of the Beijing Auto Show, although the pursuit of technology by intelligent driving suppliers is not the same, the plan to expedite the implementation of high-end intelligent driving in 2024 has been very consistent.
In addition, according to the cost of the solutions given by various suppliers at this exhibition, the price of high-end intelligent driving in the 100,000~200,000 models accounts for only 3%~5%.
This also means that under the technical maturity and intelligent price involution of high-end intelligent driving, the vast majority of smart cars have the opportunity to land high-end intelligent driving in 2024.
In anticipation of the completion of the intelligent transformation of the car at the travel level, consumers have begun to pay more attention to the intelligent experience of life on the car.
The large model "gets on the car", and the intelligent interaction of the cockpit is more intelligent
The trend of large models landing in intelligent cockpits has long been a consensus among car companies around the world.
Overseas, at CES 2024, Volkswagen, Mercedes-Benz, and BMW have used large models to create cockpit personal assistants. In China, it is the ideal Mind GPT, NIO's NOMI GPT, the cockpit model of the Volcano engine, and the automotive industry model of Tencent's smart mobility.
Wang Xiaogang, chief scientist of SenseTime, believes that "after the large model is 'on the car', it can combine the single-point AI functions of the cockpit and automatically call the cockpit software and hardware resources, and the multi-modal large model can realize the multi-sensory interaction between people and vehicles, and significantly improve the smoothness and naturalness of interaction."
At this year's Beijing Auto Show, suppliers of automotive intelligent cockpits gave some of the latest thoughts on the "getting on the car" of large models.
Among them, SenseTime gave the "most sci-fi" answer to a large-scale model landing cockpit in a way similar to Apple's Vision Pro's eye and gesture interaction in the intelligent cockpit.
SenseTime's ability to interact with eyes comes from the accumulation of DMS (driver monitoring system, which can recognize fatigue driving) in the car. With the improvement of recognition accuracy, SenseTime has realized the precise gaze interaction of the driver's central control screen or the rear screen of the rear user.
The interaction method of "see where to point" is very ideal, so that the driver can control the car machine during driving without reaching out to touch the car screen, which improves the safety of driving. However, the accuracy of the specific operation will be shared with you after the follow-up experience and evaluation.
On the other hand, in terms of 3D gesture interaction, SenseTime has also designed an end-to-end recognition network for gesture recognition on the basis of a large amount of real person data, so that users can realize the car machine operation that previously needed to be touched "in the air". This makes it easier for users in the back to control the front screen, control music playback, switch maps, and more.
From the new interaction technology, we can see that the car has created a more three-dimensional operation space for the landing of AI, and the interactivity has drawn a lot of essence from PCs and mobile phones, and it has also opened up differentiation in practicality.
In addition to interaction, the landing of the cockpit model is also a highlight of this year's large model.
This year, Tencent Smart Mobility has upgraded a new cockpit model based on the hybrid model, which can do intelligent Q&A of the car's AI manual, such as the professional car/car maintenance Q&A for the model, and actively provide vehicle condition reminders and maintenance suggestions; based on the App Agent capability, it only takes 1 week to learn and skillfully use 1000+ in-vehicle applications, realize in-depth voice interaction of in-vehicle applets/APP, and help users complete various in-vehicle service operations, such as automatic coffee ordering, movie ticket booking, Booking valet service and other operations.
In the Qiankun intelligent car solution, based on the Pangu model, Huawei has trained its own large model of the automotive industry, the Qianwu engine, and with the blessing of this AI engine, the HarmonyOS cockpit has been iteratively upgraded in the car machine, audio, display and other links.
Huawei Qianwu Engine
At the in-vehicle level, based on Huawei's self-developed HarmonyOS system, Huawei's in-vehicle machine can achieve smooth multi-device connection while also supporting the native loading of most commonly used apps. Compared with the "few" apps on the car machine of other car companies, everything in the car machine has to be projected on the mobile phone, and the Hongmeng car machine can be called "far ahead" in the application ecology.
Huawei has a huge number of car machine APPs, and the model is AVATR 12
In terms of sound and display functions, Huawei also uses AI to adapt to the video, automatically switching the video sound to 3D sound effects. The user is able to accompany the picture and hear the sound from all sides of the picture. In addition, Huawei also uses AI to carry out "directional noise reduction" in the cockpit space, so that the front row listens to songs and the back row plays games without disturbing each other.
In addition to Tencent, Huawei and SenseTime, OEMs are also trying more possibilities of intelligent cockpits at the level of interaction and experience 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 Motor's intelligent cockpit also realizes an interactive technology similar to SenseTime's gesture control, and users can complete a series of operations such as playback, switching, and selecting through specific actions such as pinching, dragging, and clenching fists. In addition, Great Wall Motors has also equipped the Coffee Life App on the Haval series models, and the car can be connected to smart peripherals such as smart water cups, child seats, and car refrigerators to achieve functions such as warming milk, making tea, and temperature control.
WEY brand Blue Mountain cockpit
In addition, the recommendation combined with the location is also a new highlight. Zhiji Auto and Volcano Engine have cooperated in the field of large-scale model applications and algorithms to create City Drive, which can filter and push local life content such as special food, popular attractions, and cultural landmarks in real time according to users' personalized preferences, so that travel can expand from "I want to go" to "take me there".
On the whole, the intelligent cockpit of automobiles and the function-based interaction mode of traditional vehicles and machines have become farther and farther away, and they have been far ahead in terms of intelligence, functions and application richness.
Fang Yunzhou, the founder of Nezha Automobile, believes that "the large model 'getting on the car' has now brought about a change from 0 to 1, and the future will achieve greater changes from 1 to 100, and people will hand over the driving to the car and deal with other things in the intelligent space."
It is foreseeable that car companies will continue to increase investment in automotive intelligence, in order to be ahead of their opponents at every stage, and the "rally" of comprehensive intelligence is about to start.
Intelligent warfare beyond the vehicle
Seeing that the "supply line" needed to build a smart car is "pulling" longer and longer.
In order to maintain long-term intelligent competitiveness, the cost of car companies is rising. The gap between the excellent and the backward is also "inflating" at a speed visible to the naked eye.
"There is even a feeling that the century-old advantage of a joint venture brand has collapsed in the era of smart cars." A senior automotive industry analyst lamented to Light Cone Intelligence, "Even if it is a new car-making force that once sold well, it has not increased investment in intelligence in the past two years, and it is immediately a little lonely at the auto show."
On the one hand, it is necessary to continue to "shop" in the tide of price reductions, and on the other hand, it is necessary to maintain intelligent and continuous expenditure. Even if car companies know that this money has to be spent, they will always take a look at the market, which is a more cost-effective solution than building a self-built intelligent computing center and spending billions more per year.
As a result, the capabilities and foundation of AI large model manufacturers (mainly cloud vendors) are also the objects of collective attention of car companies.
With the combination of automotive intelligence and AI cloud, with the landing and continuous iteration of high-end intelligent driving, the person in charge of intelligent driving from a car company knows better how to mobilize this demand and anxiety.
Wu Xinzhou, global vice president of NVIDIA, said, "After entering the era of AI-defined automobiles, the development model of automotive intelligent driving has changed, forming a closed loop of cloud + vehicle development - intelligent driving models are trained and verified in the cloud with large computing power, and then deployed to the vehicle to complete application and data collection and backhaul, and promote model training in the cloud."
For example, Huawei's intelligent vehicle solution Qiankun 3.0 learns 30 million kilometers per day, and the model can be iterated every five days. At this speed, it's really terrifying.
In addition to the training of intelligent driving, the R&D, marketing, and office of car companies all need large models to accelerate digital intelligence.
For example, Tencent's large model of the automotive industry can assist software engineers in writing code in the R&D process, intelligent customer service and sales AI assistants in the marketing process, and even build an AI engineering platform that car companies can directly use to develop the AI applications they need.
In the final analysis, the current car companies continue to increase investment because smart cars are still a long way from the "ultimate product". Perhaps in a few years, car companies will be in the midst of price reduction and intelligent investment. However, it is gratifying that Chinese auto brands will also build insurmountable barriers on the road of continuous pursuit of intelligent technology.
#智能汽车#