The secret to competing in intelligent driving lies in the latest technology trends

Recently, the China Society of Automotive Engineers released the "Top Ten Technology Trends for China's Automobiles in 2025" based on forward-looking predictions and realistic considerations of my country's automotive technology innovation directions, focusing on sub-sectors such as new energy, intelligent networking and intelligent manufacturing.

Judging from the "Top Ten Technology Trends in China's Automobile Industry in 2025" (hereinafter referred to as the "Top Ten Technology Trends") selected this time, content related to intelligent driving occupies half of the list, including multimodal large models, on-board intelligent computing platforms, AI-enabled synthetic data, safety risk management systems, and cross-domain integration of intelligent chassis and other cutting-edge technologies.

At present, the second half of the "intelligent" battle is in full swing. The research and development, mass production and large-scale application of the above technologies are the top priorities of this year's intelligent driving competition, and are also the battlegrounds for the new intelligent driving battlefield in 2025. With the latest intelligent driving technology as the foundation, the future automotive industry will usher in a more intelligent development pattern.

In addition to the currently popular intelligent driving technology, the latest trends in other sectors are also selected, for example: the power consumption of new energy A-class passenger cars per 100 kilometers will be reduced to below 10kWh, intelligent and efficient hybrid control strategies will continue to be optimized and become increasingly popular, and EMB technology will become increasingly mature and will soon be ushered in mass production applications.

However, as the core track of new energy vehicle technology innovation and investment and financing, each major technology of intelligent driving determines the subsequent transformation and commercialization of the industrial chain. With the iterative development of big computing power, big data, and AI big models, high-level intelligent driving will also achieve revolutionary breakthroughs and promote intelligent connected vehicles into a new stage of development.

01

In-vehicle intelligent computing platform reduces costs and improves quality

Help NOA and other intelligent driving technologies develop rapidly

The on-board intelligent computing platform provides core computing power support for intelligent driving and is an important indicator of the level of automobile intelligence.

At present, the performance difference of in-vehicle intelligent computing platforms has become a core factor affecting the level of intelligent driving. Specifically, the intelligent computing platform integrates multiple SoCs, supports large amounts of data parallel computing and complex logic functions, and has higher computing power and energy efficiency; through the coordinated optimization of software and hardware, it can achieve more efficient computing power utilization, reduce power consumption and improve system stability.

According to statistics from the "Top Ten Technology Trends", the current in-vehicle intelligent computing platform has achieved ultra-large storage bandwidth, supports efficient data instruction exchange, and has a computing power of up to 500 TOPS or more, which can meet the demand for higher computing power caused by the growth of massive data in advanced models such as end-to-end.

In 2025, the performance of in-vehicle intelligent computing platforms will continue to upgrade, and the cost will show a trend of further reduction, which will help NOA and other intelligent driving technologies to have a penetration rate close to 20%. In recent years, domestic and foreign manufacturers such as Mobileye, Horizon Robotics, Huawei, and NVIDIA have actively deployed the mass production of in-vehicle intelligent computing platforms.

Take Horizon Robotics as an example. The company's independently developed intelligent computing architecture BPU (Brain Processing Unit) simplifies the development process, improves the intelligence and efficiency of computing, and reduces the additional costs and performance losses caused by software and hardware adaptation issues.

At the software level, in order to implement the end-to-end evolution concept of autonomous driving, Horizon has applied technological innovations and breakthroughs at the algorithm level to its latest intelligent driving solution, SuperDrive, which can more efficiently process complex traffic scene information and enhance the perception and decision-making capabilities of the intelligent driving system.

02

Integration of intelligent driving and intelligent chassis

Improve the motion control performance of vehicles above L3

In the past few years, smart driving and smart cockpits have become the hottest topics in the industry, but many people tend to overlook that chassis technology, as a core role in the intelligent process, is no less important than smart cockpits and smart driving.

The deep integration of intelligent driving and chassis has become the key cornerstone for making autonomous driving cars safer and more efficient.

这里说的深度融合，主要是通过传感器、执行器和控制策略的深度融合，增强自动驾驶汽车的感知和决策能力、精准控制能力显著提升行车安全、驾驶舒适性与系统效率优化，降低自驾退出频次，助力L3以上自动驾驶落地应用。

The deep integration of intelligent driving and chassis involves multiple aspects such as system integration, intra-domain integration, and cross-domain integration. It includes key technologies such as enhanced chassis perception technology, wire control technology, and redundant design of collaborative control algorithms, which will reshape the transformation of the whole-partner cooperation model in the automotive industry.

The "Top Ten Technology Trends" analyzes that by 2025, through the deep integration of intelligent driving and intelligent chassis, major breakthroughs in chassis intelligent motion control technology under extreme working conditions will be achieved, such as automatic non-stop parking function linked with intelligent driving and power, energy consumption control recovery system and magic carpet function.

Last month, Chery officially announced that it had joined hands with Huawei to release the Gimbal Intelligent Chassis 2.0 technology, which adopts a new generation of electrical architecture and 1000 TOPS computing power. The OTA upgrade time of the entire vehicle can be less than 25 minutes.

The Zhijie S7 is the first model to be equipped with Huawei's Tuling chassis. During the pre-sale stage of the new car, the intelligent chassis was one of the highlights and was promoted by Huawei. Through digital capability upgrades, Huawei's Tuling chassis connects the chassis, cockpit and intelligent driving, which can better coordinate and control the entire vehicle.

In addition, NIO has completed the technical advancement from building an intelligent chassis with NT1, to AI empowerment with NT2, and then to the Tianxing chassis with NT3, leading the further technical exploration of the intelligent chassis through the full-wire active suspension.

It is reported that one of the "trump cards" of NIO's AI smart chassis is NIO's full-domain operating system for the entire vehicle - "SkyOS Tianshu", which connects the intelligent driving, cockpit, vehicle control, vehicle networking and other domains at the bottom layer, and can realize the free call of the computing power of the four Orin intelligent driving chips to deploy the whole vehicle algorithm.

03

AI empowers synthetic data

Will become an important resource for autonomous driving research and development

Using advanced artificial intelligence technologies such as generative AI and world models to generate high-quality synthetic data can effectively alleviate the problem of data shortage and improve the reliability of algorithm models. This is a very promising development direction for autonomous driving model training.

In the field of intelligent driving, high-quality real data is becoming an increasingly scarce resource. The advantages of synthetic data are low collection cost, self-annotation, strong cross-platform versatility, and the ability to supplement potential dangerous scenarios and edge scenarios in a targeted manner, thus improving the long-tail scenario library.

Advanced AI technology can process different types of data such as text, images, and videos, and can quickly extract valuable information from large amounts of unlabeled data, such as information about different types of traffic participants and interactive behaviors between traffic participants.

The "Top Ten Technology Trends" believes that by 2025, synthetic data generated by AI will be widely used for efficient training and simulation services of autonomous driving models.

In 2024, companies such as Tesla, Nvidia, Wayve, Baidu and NIO have taken the lead in deploying forward-looking research and development and application of world models, supporting the use of some algorithms in vehicles.

As the pace of high-level autonomous driving accelerates, AI synthetic data will gradually replace traditional data collection methods in 2025 and become the way to meet the data needs of autonomous driving.

04

Multimodal large model

Promoting innovative breakthroughs in autonomous driving perception and decision-making

The "Top Ten Technology Trends" points out that by 2025, autonomous driving models are expected to achieve important breakthroughs in autonomous driving perception and decision-making capabilities as multimodal large model algorithms are improved, data generation capabilities are enhanced, computing power and training time are expanded.

利用多模态大模型通识能力，可有效应对智能感知中存在的长尾问题，显著改善车辆对场景、障碍物、导航信息等要素的理解能力。