(Image source: Unsplash)
The 2024 Beijing International Automobile Exhibition has come to an end.
According to the data, during the Beijing Auto Show, 117 new models made their world premiere, and about 280 all-electric, petrol-electric hybrid, trucks and other types of vehicles were exhibited, and more than 500 car factories and suppliers such as Xiaomi Automobile, Volkswagen, Nissan, and SemiDrive Technology displayed their products. A total of 892,000 visitors poured into the show in 10 days, of which 3% were dealers.
One of the biggest highlights of this year's exhibition is the "car" of China's AI model.
During the auto show, Nissan Motor announced that it had signed a memorandum of understanding with Baidu to conduct feasibility studies on strategic cooperation in the field of AI and smart cars. Toyota, Changan, VOYAH, GAC and other vehicle brands announced that they have reached a cooperation with Tencent to jointly create a large-scale model solution for the automotive industry, covering scenarios such as cockpit assistant, marketing, and customer service. Previously, SenseTime and Xiaomi announced a cooperation to equip Xiaomi SU7 with RiRixin large model technology in the cockpit of the car.
According to incomplete statistics compiled by Titanium Media App, more than 20 auto brands have announced that AI large models are officially "on the car".
The concept of large models "getting on the car" is becoming more and more popular, especially many Chinese AI companies are "standing out of the encirclement" and competing for international brand car cooperation.
With the quiet arrival of the second half of smart cars, cars are being upgraded to a new generation of "mobile intelligent terminals", and their intelligence level has also become the key to the success of car companies in the era of intelligent networking. Accelerating the "car" of large models has also become a new choice and direction for car companies to highlight product intelligence and create product differentiation.
However, the editors of Titanium Media App also found another phenomenon at the station site, that is, the large model "on the train" has not yet been commercialized on a large scale, and the feedback from many companies is in the "display" stage. Some experts pointed out that at present, the concept of large model on the car is the majority, and the specific practical application remains to be seen, and the large model on the car may not be able to improve the business performance of enterprises and bring stable profits.
Hong Wanting, a senior analyst at IDC China, told Titanium Media AGI that intelligent driving technology is essentially a complex system engineering with AI as the core driving force, and its development is highly dependent on the company's technology research and development background, capital investment scale, and keen capture and efficient transformation capabilities of cutting-edge technology. The continuous cooperation between international car companies and domestic enterprises in the field of intelligent assisted driving or AI is an important step for China's intelligent driving technology to stand on the world stage, and the core driving force behind it is the in-depth layout and continuous investment of domestic enterprises in AI technology.
The AI model is going to be "on the car", and car manufacturers will leverage the 1.4 trillion market
According to the latest data from the Passenger Car Association, from April 1 to April 14, the national passenger car market retailed 516,000 units, of which the sales of new energy vehicles increased by 32% year-on-year to 260,000 units, with a penetration rate of 50.39%; Passenger car manufacturers in China wholesaled 534,000 vehicles, including 268,000 new energy vehicles, with a penetration rate of 50.19%.
The retail and wholesale penetration rates of new energy vehicles both exceeded 50%. Earlier, the penetration rate of electric vehicles in new car sales has increased rapidly, reaching about 35.3% from 2021 to May 2023, and the electric vehicle market has gradually entered a stage of stable and rapid growth from the explosive growth stage.
At the same time, after many years of competition in the fields of appearance, electric and intelligent cockpit, deep intelligence has become the next target anchored by Chinese automobile companies.
The China Association of Automobile Manufacturers predicts that by 2035, the scale of China's smart car industry will exceed 200 billion US dollars (about 1.44 trillion yuan).
The UBS UBS research team predicts that by 2025, China will produce as many as 6.6 million new energy vehicles, that is, one out of every four new energy vehicles will have intelligent functions; IDC believes that by 2026, the global sales of intelligent driving vehicles will reach 89.3 million.
In fact, the initial connection between AI and automobiles dates back 40 years, and the initial direction was "autonomous driving".
In 1984, the U.S. Defense Advanced Research Projects Agency (DARPA) partnered with the Army to launch the Autonomous Ground Vehicle (ALV) program. Subsequently, Carnegie Mellon University, Stanford University, Massachusetts Institute of Technology and other American universities have joined the research work on driverless cars.
In 1995, NavLab-5 achieved an average speed of 88.5km/h in the experimental environment, and crossed the continental United States for the first time, with an autonomous driving mileage of 4496km, accounting for 98.1% of the total trip.
In October 2015, AI technology and automobiles ushered in an important point in time - Tesla launched the semi-autonomous driving system Autopilot, becoming the first manufacturer in the world to put into commercial autonomous driving technology. Subsequently, General Motors launched Cruise, and the Audi A8 mass-produced Level 3 autonomous driving technology.
At the same time, at the domestic level, China's driverless car research is later than that of the United States. In 1992, the National University of Defense Technology successfully developed China's first truly driverless car.
In July 2011, the Hongqi HQ3 driverless car, jointly developed by FAW Group and National University of Defense Technology, completed a 286km driverless test for highways.
By 2015, Yutong large buses completed autonomous driving tests in a completely open road environment, which was also the first automatic driving test of large buses on highways in China; At the 2016 Beijing Auto Show, BAIC Group showcased its driverless car based on the EU260, which became an unmanned perception and control car.
Nowadays, electrification, networking, and intelligence have become the development trend of the automobile industry.
So, before the birth of large models, why didn't AI "get on the car"? The editors of Titanium Media AGI believe that there are three points:
First of all, it is difficult for new energy vehicles to achieve "autonomous driving" in the short term. From the perspective of technical experience, autonomous driving is a kind of driver assistance from the starting point to the end point, between the "two points" distance using vision, lidar, Internet of vehicles and other technologies to achieve driver assistance or its completely uncontrolled car, therefore, in view of China's traffic jams in Beijing, Shanghai, Guangzhou and Shenzhen, maps can not be updated in real time, policies and regulations and other factors, China has been designing autonomous driving as experimental and regional, such as Beijing Yizhuang, Beijing Shougang, etc. Therefore, the "autonomous driving" that everyone was expecting before could not be implemented, and the urban NOA was only an assist, or a technical "balance" made by using high-definition maps, rather than real autonomous driving.
Second, traditional AI uses technical solutions such as decision trees and unsupervised learning to eventually approach a correct value when the amount of data cannot reach the scale level. Of course, this is also due to the fact that end-to-end scenarios such as automobiles cannot have large, 100% valid high-quality datasets. In addition, traditional AI cannot achieve general-purpose scenario technology.
Finally, the computing power is not strong enough, and the capital and labor costs are too high. He Xiaopeng, chairman of Xiaopeng Motors, once said that the research and development of automotive intelligence belongs to large computing power and big data projects, which require a lot of capital and manpower, and it is extremely difficult for general computing power companies to do. Prior to this, from 2017 to 2021, new energy vehicles, AI, cloud computing and other fields were all striving for revenue, commercialization, and capitalization.
Zhou Hongyi, chairman of 360 Company, told Titanium Media AGI that AI was previously "born" intelligent cockpit and autonomous driving on the car, but it was also hindered by the technical route and was not "true artificial intelligence". But now, AI large models have brought revolutionary changes, and the "getting on the car" of large models is a fundamental improvement in the intelligence ability of the car.
Zhou Hongyi gave four cases of the scene of the "getting on the car" of the big model to AGI for titanium media.
- First, the previous intelligent cockpit, the user dialogue is more based on the rules processing, you say open the window, he can understand, you want to say shake the glass down, he can't understand, similar to Siri, in fact, it is a mentally retarded character, not easy to use. Nowadays, after the large model is in the car, it is a revolutionary thing to really let the car understand your thoughts.
- Second, Sora's current capabilities represent multimodal capabilities, and in the future, automobiles and robots can use sensors such as lidar and cameras to achieve perception, cognition and other capabilities. At the same time, this multi-modal large model can make the car better understand many things inside and outside the car, such as distinguishing whether it is the owner of the car, whether it is fatigue driving, etc.
- Third, today's car is still a mechanical object, it can't understand many things, it can't do logical reasoning and planning, including robots, it is the same as receiving human instructions. However, in the future, once a large model is added to the car, the large model will have the ability of the brain, which can not only understand human discourse communication, but also fully understand human knowledge and reasoning, analyze the process of logical judgment, and make more judgments.
- Fourth, autonomous driving will make a breakthrough in the next two years. Transformer's large model training is different from traditional intelligent driving, traditional intelligence is still a rule-based scheme, mainly to change bugs and software upgrades, then there are 10,000 rules, what should I do when I encounter this situation, there are always cases that are not edited into the rules. If the Transformer is used for training, it can provide a variety of driving cases. The most powerful ability of the large model is that it does not send you into the data storage, but trains it through your data, and it can form a summary of the inferences.
Nowadays, the intelligent "car" of AI large models has become a new trend at this Beijing Auto Show.
As a leading Japanese auto brand, Nissan (China) Investment Co., Ltd. and Baidu Online Network Technology (Beijing) Co., Ltd. jointly reached a strategic cooperation memorandum of understanding on the feasibility study on the first day of the auto show.
Nissan revealed that the cooperation content of this feasibility study includes two major aspects: first, based on Nissan's intelligent platform, Baidu's advantages in generative AI, jointly develop and study the future technology development and commercial cooperation feasibility in related fields; The second is to equip Nissan with Baidu's AI solutions in Chinese models to explore the development potential of AI solutions and jointly explore the application and development of innovative technologies in the future automotive field.
Makoto Uchida, president and CEO of Nissan Motor Company, told Titanium Media App that China's new energy vehicles are developing very rapidly. The main focus of this collaboration is to better integrate these technologies into Nissan's products to bring exciting intelligent systems and AI-based services to Chinese users.
Masashi Matsuyama, Vice President of Strategy and Special Missions of Nissan Motor Company China and General Manager of Nissan (China) Investment Co., Ltd., said, "The automotive industry is undergoing intelligent transformation, and consumer demand for new technology applications represented by AI is growing. Through the partnership with Baidu, Baidu's generative AI capabilities will be applied to Nissan's vehicles in China, which will further accelerate the pace of Nissan's intelligent and electrified vehicles, and provide consumers with a more diversified smart travel experience. ”
Masashi Matsuyama emphasized that Nissan's cooperation is not limited to the electrification of vehicles, but also includes intelligence.
Obviously, deep "intelligence" is not only the next goal anchored by Chinese auto companies, but also an important development direction for global auto leaders.
The practical application remains to be seen, and the large model "getting on the car" still needs patience
Zhou Hongyi bluntly said to the titanium media app that although the "car" of the large model will bring "revolutionary" changes to the automobile industry, this change will not happen "overnight".
(Source: Edited and photographed by Titanium Media App)
Titanium Media App found at the Beijing Auto Show that although car brands are promoting the concept of "AI large model on the car", the AI large model "on the car" is still in a PPT state at the auto show site, and most of the unveiled show cars have set up the show car mode in consideration of safety, and there is no way to directly experience the real scene effect of the AI large model on the spot.
Tang Daosheng, senior executive vice president of Tencent Group and CEO of the Cloud and Smart Industry Business Group, believes that the "getting on the car" of large models requires patience. Upgrade key capabilities such as computing power, algorithms, and data to promote breakthroughs in autonomous driving. AI has become the core driving force for user experience upgrades. The cycle of technology iteration and product update in the automotive industry continues to shorten, and it also needs matching operation management capabilities and industrial chain collaboration capabilities.
When the entire automotive industry is caught in the vortex of price cuts and concessions, there is uncertainty about how many companies will be able to win the competition in the big model car. At the Beijing Auto Show, many auto brands did not disclose the mass production time of the latest large-scale model products.
However, in the long run, the "onboarding" of intelligent technologies, including large AI models, will definitely become the most important trend in the automotive industry.
At present, Chang'an, Geely, Lantu, Hongqi, Great Wall, Dongfeng Nissan, Leap, etc. have all taken Baidu's "express" of Wenxin Yiyan, focusing on the improvement of voice interaction; Not long ago, Geely and Baidu joined hands again, and the high-end intelligent car robot brand - Jiyue was born, seeking a breakthrough in intelligence; Earlier this month, Huawei released the HarmonyOS 4 system that is connected to the Pangu model, which will be the first Huawei smart model Luxeed in cooperation with Chery.
Lei Jun said that smart electric vehicles have been developed for 10 years, in fact, the main solution today is the "electric problem". The past 10 years have been a decade of electrification, and the "second half" that begins today is a decade of intelligence. Therefore, intelligence is the core and soul of future intelligent electric vehicles.
Wang Xiaogang, co-founder and chief scientist of SenseTime, and president of Jueying Intelligent Vehicle Business Group, recently told Titanium Media App that the AI model "getting on the car" essentially changes two aspects: one is the improvement of production efficiency, and the second is the change of human-computer interaction.
Wang Xiaogang mentioned that the manpower required for automobiles is very high, and in the future, large-scale models and machine automation may save 80% of the workload. In more irregular and complex scenarios, large models will adapt to more needs in many human-computer interaction scenarios such as autonomous driving and intelligent cockpits.
"The scene itself is very important, and today, smart cars are going through the process from automotive intelligence 1.0 to general artificial intelligence." Wang Xiaogang emphasized that in the future, large-scale model technology can also be extended to the field of robotics.
It is reported that by the end of 2023, SenseTime has reached cooperation with more than 30 domestic and foreign car companies such as Honda, BYD, Great Wall, GAC, Hongqi, and NIO, with AI solutions covering more than 90 models and a total of 1.9 million smart cars delivered.
Hong Wanting told Titanium Media App that in just a few years, we have seen that the intelligent driving function of Chinese car companies has been able to solve challenges such as unstable driving trajectory, frequent sudden braking, and only being able to turn on at a certain speed, and has begun to be able to better complete end-to-end driving tasks in high-speed and urban areas.
Looking to the future, the role of AI models in the automotive industry may not only be in a single field such as intelligent cockpit and intelligent driving, but also in generalized scenarios. Of course, there is still a long process of exploration and testing that will take place.
Li Keqiang, academician of the Chinese Academy of Engineering, professor of Tsinghua University, and chief scientist of the National Intelligent Connected Vehicle Innovation Center, said that at present, the application of large models in the automotive field is still in the early stage of trying and exploring.
At present, there are still many risks in the application of large models in the automotive field: on the one hand, relevant policies and regulations have not yet been promulgated, data security cannot be guaranteed, and data compliance has become the key; On the other hand, the landing scenario is not yet clear, and this is where the challenge lies.
With the continuous change of AI large model technology, it remains to be seen what "changes" will occur in the automotive field in the future, and we need to continue to wait for important changes in technology.
(This article was first published on Titanium Media App, author | Lin Zhijia, editor | Hu Runfeng)