“The first half is electrification, the second half is intelligence.” Following the development trends of connectivity, electrification, and greening, the competitive landscape of the automotive industry has rapidly embraced a new focus — intelligence. Market analysis firm McKinsey predicts that by 2030, China’s passenger car market will expand to a massive scale of over 300 million vehicles. Artificial intelligence technology plays the most critical role in this process and is expected to create an enormous economic value of over 380 billion US dollars.
With the “AI in ALL” trend sweeping through the automotive industry at a rapid pace, well-known car manufacturers, including Red Flag, Great Wall, Dongfeng Nissan, and Geely, have invested in the “Wenxin Yiyan” ecosystem, showing a strong ambition to master emerging technologies. By applying AI big models, these companies are not only tackling the digital age but also undertaking an important layout aimed at creating competitive differentiation.
According to data from the China Association of Automobile Manufacturers, in 2023, China’s commercial vehicle production and sales volumes reached 4.037 million and 4.031 million, respectively, with year-on-year growths of 26.8% and 22.1%, outpacing the industry’s average level. The continuously expanding market, coupled with an increasing number of domestic brands joining the competition, and the evolving demands of consumers, indicate that the automotive industry is set to transcend the tipping point of relying solely on pricing strategies, moving towards a new phase of competition centered on technology and intelligence.
The application of AI big models has become the driving force for this transformation, leading the new competitive focus on the implementation of intelligence and comprehensive digitization of vehicles. Automotive companies are leveraging AI technology to optimize design processes, improve production efficiency, and enhance customer experience, which is the core competitive strategy of the new era of technological intelligence.
At present, as more and more companies begin to integrate big models into their operating systems, we are witnessing a reality — how intelligent car manufacturing is influencing the future direction of the whole industry.
Big Models Take the Wheel
(I) Intelligent Driving VS Autonomous Driving
The rise of big models has brought new vitality to the development of autonomous driving technology. The core of this technology lies in accurately recognizing and judging environmental information collected through numerous sensors. With their massive data analysis capabilities, multi-dimensional processing abilities, and comprehensive predictive powers, big models are indispensable in solving the challenges of data annotation and other issues in autonomous driving.
In April 2023, Great Wall Motor’s subsidiary Haval Holo released the world’s first generative big model for autonomous driving development, DriveGPT Xuehu·Hairuo. The new model employs RLHF (Reinforcement Learning with Human Feedback) technology and real human driving takeover data to continuously optimize the decision-making cognition model for autonomous driving. DriveGPT has covered 40 million kilometers of driving data in training, with a parameter scale reaching 120 billion. Although it has not yet achieved end-to-end autonomous driving, it is gradually advancing to the stage of integrating perception, cognition, and control models.
Compared to the concepts of unmanned driving and full autonomous driving, the Navigation on Autopilot (NOA) technology used by SAIC Group’s Zhiji Motor seems closer to reality. Zhiji Motor collaborated with the world’s top intelligent driving algorithm company, Momenta, to launch the industry’s first AI model based on D.L.P. (Deep Learning Algorithm), capable of modelling the three key stages of perception, fusion, and prediction, and achieved deep integration.
On April 8 this year, a revolutionary new car, the Wisdom L6, made its debut, equipped with the DDOD (Data Driven Object Detection) model and the DDLD (Data Driven Landmark Detection) perception big model. The combination of these two models enables the monitoring and recognition of dynamic objects on the road, as well as the identification of road surfaces and static elements, which greatly improves the precision of autonomous driving. In addition, the highly anticipated NOA driverless city navigation system is expected to launch this year, extending the technology nationwide.
While fully automatic driving remains in the near future, innovations in smart cockpit technology continue to enrich the driving experience, meeting market demand and pushing the development of cars towards a “mobile third space”. Last June, Wisdom Motors released its car-wide intelligent software, “Full Journey AI Cockpit”, which integrates advanced software and hardware technologies, with significant enhancements in safety and comfort. Meanwhile, Chery Automobile and GAC Group have launched the “LION AI” and AI big model platforms, respectively, both achieving substantial progress in intelligent voice interaction, providing a more natural voice dialogue experience.
Furthermore, Geely Automobile inventively introduced an exterior AI voice interaction feature that can recognize voice commands to perform various operations, such as opening the trunk or adjusting the climate control and even implements greeting functions upon entering and leaving the vehicle. Geely’s Xingrui AI big model integrates numerous AI-native applications, such as AI picture books, AI memories, AI music rhythm, and others, enhancing the immersive experience in-vehicle.
At the AI big model application level, companies like BYD and BAIC BluePark are accelerating the thorough intelligence of cars. BYD’s dual-loop multimodal AI platform “Xuanji” covers over 300 usage scenarios, and by integrating the intelligent architecture “Xuanji”, it breaks down barriers between different systems. BAIC’s EXEED just launched the “Darwin 2.0” technology system, emphasizing the importance of intelligence, device collaboration, and information sharing. Through the system’s self-evolution, it minimizes human intervention and enhances the efficiency and safety of the vehicle.
Yang Jifeng, the head of Great Wall Motors AI Lab, points out that many OEMs are still competing over basic intelligent functions, such as voice control, DMS, ambient lights, etc. However, these should not be considered true AI issues but merely scene definitions. In the future, when the smart cockpit evolves into a comprehensive intelligent space, it will be a real challenge for AI technology. Such intelligent spaces will combine multimodal perception, cognitive big models, and AIGC big models, significantly enhancing overall AI capabilities, based on strong data support and deep algorithmic reasoning, to achieve more natural human-computer interaction. While this concept is attractive, the path to realization remains long and challenging.
As the automotive industry continues to move towards digitalization, the role of data has transformed from a secondary by-product of production to a core element of production.The application of AI large-scale models has connected the data between different stages such as production, design research and development, financial management, sales, and after-sales service, promoting smooth interoperability of data across the entire supply chain. Such applications are not only to facilitate smoother car sales, but their deeper significance lies in that they have reshaped the way cars are produced, and essentially achieved the goal of cost reduction and efficiency enhancement.
An executive in the automotive industry once expressed the view that the core of enterprise transformation is the shift from a traditional reliance on processes and responsibilities to a data-centric operating model, which must respond rapidly to customer needs, forming a continuously evolving business and development capability. Moreover, a global partner at a consulting firm pointed out that, although the complexity of in-car software has more than doubled over the past decade, the improvement in software development efficiency is only between 1 and 1.5 times. With the rapid changes in the market, shortening the R&D cycle, reducing the development threshold, and improving R&D efficiency have become key for car companies to maintain competitiveness.
In this context, China FAW has attempted to optimize its business processes using AI large-scale models. This year, they developed and implemented a business intelligence application based on large-scale models with their partners. This application brought disruptive changes to business processes through automated report generation and decision support. FAW even used large-scale models to write code, achieving automation in design and auto-generation of code. Such large-scale model systems will continue to iterate. Currently, AI large-scale models can undertake half of the coding work, it was reported.
Another case is the Star Rui AI large-scale model adopted by Geely Automobile. It is the result of the deep integration of the company’s self-developed natural language processing model with the R&D system, supporting applications in domains such as modeling, mechanical design, quality control, and autonomous driving. The large-scale model has helped Geely save time and cost in R&D, significantly improving efficiency.
However, as the application of large-scale models brings convenience, car companies also face challenges in technology and infrastructure. For example, due to limited on-board computing power, many AI processes still rely on cloud services. To this end, some companies have established their intelligent computing centers to ensure back-end computational support.
At the same time, although a large amount of data provides a basis for AI algorithm development, how to efficiently collect, clean, and train data from different scenarios and dimensions remains a challenge. And whether the application of large-scale models indeed brings cost reduction, as well as the actual effect of the massive R&D investment by many car companies, remains inconclusive.
The challenges faced by companies participating in the artificial intelligence (AI) competition for smart cars are extremely severe, especially in terms of the cost of smart driving chips. Over the past three years, the self-sufficiency rate of Chinese automotive chips has significantly increased, jumping from 5% to 10%, and with local suppliers such as Horizon Robotics and Black Sesame Intelligence growing rapidly. However, overall, the onboard chip market is still dominated by foreign brands.
For the various components of smart vehicles, procuring from multiple suppliers makes it very difficult to integrate and coordinate different systems, as well as reduce R&D and production costs. Not to mention realizing mass production on a larger scale. From this perspective, BYD continues to follow a full-stack strategy for independently developing smart vehicle technology, which seems to be a wise move.
Faced with the industry impact brought about by the AI wave, whether it is passive or active participation, the restructuring and reshuffling of the industry structure are inevitable. This is both an opportunity and a challenge, and automotive enterprises must explore firsthand.