Since the beginning of this year, "AI-driven," "end-to-end," and "large models" have become high-frequency terms in automotive intelligence. In the two major strongholds of intelligent driving and smart cockpits, manufacturers and suppliers have shown their prowess, engaging in fierce competition around AI large models, indicating that automotive intelligence has entered a period of profound transformation.

"Starting next year, leading Chinese manufacturers will increase their investment in autonomous driving, especially in the layout of high computing power of 100 eTOPS. High-speed navigation NOA, urban memory driving, and automatic parking will become standard features for models priced between 100,000 and 150,000 yuan, and we will see an explosive growth next year," said Wu Yongqiao, President of Bosch Intelligent Driving and Control Systems Division in China, at the Global Smart Car Industry Conference high-level forum on September 29. He added that if the first half of the automotive industry was about electrification and hybrid power, the second half is about intelligence. "Starting from the first half of next year, if any company cannot keep up with the pace of intermediate autonomous driving development and their models lack features such as high-speed navigation, urban memory driving, and automatic parking, they may find it difficult to enter the new round of competition and elimination."

Interviews with journalists have revealed that the integration of large models into vehicles has brought a lot of room for imagination in intelligent driving. Gao Feng, CEO and co-founder of Ouyi Semiconductor, believes that by 2030, L4-level and above autonomous driving may be commercially realized in closed scenarios and specific commercial and logistics fields. In the nearly five years from now to 2030, L2 and L2+ level intelligent driving will see large-scale popularization.

AI Becomes a New Strategic Fulfillment Point

Advertisement

In the field of new energy vehicles, electrification is referred to as the "first half," and intelligence is referred to as the "second half." In the era of artificial intelligence, the automotive industry is accelerating into a new stage where intelligence is the core competitive strength, and AI is becoming the new strategic fulcrum for the development of the automotive industry.

Wang Lang, Deputy General Manager of Chery Automobile Co., Ltd., believes that as cars upgrade from transportation tools to a new generation of mobile intelligent terminals, AI has not only redefined the car but also reshaped the management model and business boundaries of the car.

"AI is the new driving force of the era and an important tool for automotive products to achieve uniqueness and differentiation. Every era has its main contradictions and its own new driving force, just as Chery firmly chose to develop its own engine 20 years ago, AI must be a significant opportunity that Chery must seize in the next 20 or 40 years," said Wang Lang.

In Wang Lang's view, the automotive industry has entered an era of comprehensive electrification and intelligence. The complexity of automotive mechanical engineering has been greatly reduced, while the complexity and requirements of chips, software, and other technologies have increased significantly. Intelligent cockpits, intelligent driving, vehicle electronic architecture, and intelligent architecture make cars more like intelligent entities that can evolve by themselves, requiring the comprehensive integration of AI large models. Through AI empowerment, it is possible to achieve a thousand different faces for a thousand different cars, realizing differentiated, unique, and even unique intelligent experiences, giving the product core competitiveness.

The industry has already reached a consensus on this. Zeng Yongguang, Vice President of Guoxin Guofa and General Manager of Guoxin Guofa New Energy Technology Co., Ltd., believes that the automotive ecosystem has undergone an unprecedented transformation, transitioning from hardware dominance to software dominance, bringing new functions, new experiences, and new feelings. As intelligence continues to deepen, software-defined vehicles will be an inevitable trend. Under this trend, the future of cars will be determined by software technology centered on AI, rather than traditional mechanical performance and hardware configurations.

Intelligent driving and smart cockpits are the two most important parts of automotive intelligence. In intelligent driving, AI enables vehicles to respond more autonomously to complex road environments through perception, decision-making, and self-learning. In smart cockpits, AI greatly enhances the driving and riding experience of users through personalized services, natural language interaction, and emotional recognition. Large models are one of the key technologies driving the rapid development of AI.According to Momenta CEO Cao Xudong, Momenta's autonomous driving model can handle complex intersections or dynamic crossing scenarios with ease, significantly enhancing driving safety and traffic efficiency. Even in extreme scenarios such as extremely narrow parking spaces at night or cul-de-sac parking spaces, it can achieve precise parking.

AI has brought more accurate and smooth voice recognition and interaction capabilities to the smart cockpit. Wu Yongqiao stated: "In terms of the cockpit, the future direction is to create an AI cockpit. The current cockpit only improves natural semantic interaction capabilities based on ChatGPT. Bosch is currently working with several leading vehicle manufacturers to study how to deploy a large single-sided model on the cockpit, making its computing power exceed 300 TOPS for autonomous driving. This will make the operation of the smart cockpit more efficient, smooth, and intelligent."

In the view of Li Tao, General Manager of Baidu's Smart Cockpit Business Department, what the automotive industry needs in the future is a new era cockpit that can understand what users think and need, and can automatically generate a global execution plan. This is the ultimate direction of the evolution of smart cockpits.

Facing computational pressure

With the development of technologies such as autonomous driving, smart cockpits, and the Internet of Vehicles, vehicles need to process a large amount of perceptual data, decision-making, and control tasks, thus the demand for computing power has increased sharply.

Lu Sinan, General Manager of the Smart Cockpit Business Department of iFLYTEK's Intelligent Automotive Business Division, pointed out that as large models including many AI technologies are implemented in vehicles, the demand for automotive intelligent services increasingly requires higher computing power, and we are facing high computational pressure.

"In the era of artificial intelligence, what car companies lack is not production capacity. Whether there are more or fewer vehicle factories does not seem to be the main contradiction in the development of the industry. What the automotive industry lacks most is the infrastructure of intelligent computing. The shortage of intelligent computing infrastructure will become the main contradiction in the accelerated development of intelligent connected vehicles." Zhang Yongwei, Deputy Chairman and Secretary-General of the China Electric Vehicle Hundred Person Association, pointed out that there is a structural shortage of domestic automotive intelligent computing power, and there is a large gap in "mature" computing power with a complete software ecosystem.

It is understood that to complete the research and development and training of end-to-end intelligent driving, the demand for intelligent computing power must reach at least 1 EFLOPS. The average computing power of current car companies is 3 EFLOPS, and the ideal computing power is 100 EFLOPS.

According to public data, by the end of 2024, the total amount of computing resources planned by the three major operators will be 53 EFLOPS. However, for a single end-to-end large model, a company needs 100 EFLOPS of computing power.

"At this stage, how to solve the demand for computing power for intelligent driving and artificial intelligence is the top priority. We must ensure that there is available computing power, pursue lower costs for available computing power, and even solve the problem of local computing power maturing from immaturity." Zhang Yongwei believes that it is necessary to accelerate the resolution of domestic immature computing power issues, enrich software and ecosystems, create mature computing power, reduce the problem of future hardware being "strangled" by computing power, and invest a lot of money in computing power. It is necessary to continue to invest and form a scale effect around data, computing power, and algorithms. At the same time, efforts should be made to build automotive intelligent computing infrastructure and strengthen the joint construction and sharing of computing power.Computational power is the key foundation of intelligence. Li Tao believes that to address the issue of insufficient computational power, one can also take an alternative approach by avoiding the waste of computational resources in product design.

"Today, many cockpit designs involve installing a Pad in the car and directly migrating mobile apps to the car's system. Currently, it is estimated in the industry that a single car carries up to 189 apps. One can imagine how difficult it is to find the desired app among 189 while driving. This not only occupies valuable computational power and memory resources of the car system but also increases the mental and cognitive burden on users during the driving experience, and may even cause accidents," said Li Tao.

The advent of the AI large model era is a catalyst for the overall intelligence of vehicles. Xiong Zhengqiao, General Manager of the Cloud R&D Center of Bo Tai's Internet of Vehicles, believes that automotive intelligence relies on computational power, data, and the collaboration of the industrial chain. How to break down technical barriers among various parties and achieve sharing is also a significant new challenge.