Industry丨The era of AI-defined cockpit is coming

With the widespread application of artificial intelligence technology , its enabling role in various industries is becoming increasingly significant. In particular, driven by generative artificial intelligence, the cabin experience is ushering in an unprecedented transformation. In this transformation, the continuous breakthroughs in key technologies such as end-to-end large-scale model installation and computing power evolution have brought new market opportunities to the smart cockpit industry and reshaped the competitive landscape.

Smart cockpit → smart space, AI capabilities become the key

At present, in order to promote the continuous progress of autonomous driving technology, major OEMs have established large AI development teams, of which about 90% of their work focuses on the field of intelligent driving .

Therefore, intelligent driving is regarded as a core AI issue, which is a consensus of the entire industry.

Since the concept of smart cockpit was proposed, it was initially just a migration process from feature phones to in-vehicle applications. At present, the competition among major OEMs in the field of smart cockpit is mainly focused on the realization of functions such as voice control, DMS and ambient lighting.

These functions are not real AI issues, but application designs based on specific scenarios.

However, as the smart cockpit evolves into a smart space, its essence will gradually turn into an AI problem.

As an emerging concept, smart space is still in the exploratory stage.

By integrating multimodal perception, cognitive big model and AIGC big model into the smart cockpit and relying on the support of data and algorithms, the overall AI capability can be significantly improved.

Therefore, smart space is essentially an AI problem.

In the field of smart cockpits, with the rapid iteration of technology, platformization has become a key trend. This means using a common platform to meet the needs of different users.

On this basis, each brand's products will be adjusted differently according to its unique market positioning, vehicle models, prices and other factors.

For example, the size of the central control screen can be adjusted according to demand, so as to achieve differentiated competition among brands while maintaining rapid platform iteration.

MediaTek's latest smart cockpit SoC released, using 3nm/4nm process

At the NVIDIA GTC conference, MediaTek announced the launch of a series of Dimensity automotive cockpit system-on- chips (SoCs) that combine AI technology, including CX-1, CY-1, CM-1, and CV-1.

These products are all compatible with NVIDIA DRIVE OS software , covering market segments from high-end luxury (CX-1) to entry-level (CV-1), and are designed to bring excellent AI cockpit experience to the next generation of smart cars .

Among them, CT-X1 adopts advanced 3nm process technology, while CT-Y1 and CT-Y0 adopt 4nm process technology.

All three products support cutting-edge on-device generative AI technology. The CT-X1 supports AI large language models with up to 13 billion parameters, and can smoothly run a variety of mainstream large language models (LLMs) and AI mapping functions (Static Diffusion) in the car .

In addition, it also supports in-vehicle voice assistants based on 3D graphical interfaces, rich multi-screen interaction and display technologies, and advanced AI safety and entertainment applications such as driver alertness monitoring.

The Dimensity Automotive Cockpit Platform integrates the Arm v9 architecture, built-in AI computing unit and edge-generative AI lightweight technology, which can more effectively utilize memory bandwidth and capacity while ensuring AI computing accuracy.

These chips not only support large language models, but can also drive multiple screens and demonstrate powerful multimedia and audio processing capabilities.

At the same time, they also have functions such as 360-degree surround view outside the vehicle, driving record and in-cabin monitoring and care, providing strong protection for road traffic safety.

In terms of AI technology, CT-X1 excels in the field of large models. The 13 billion parameter AI large language model it supports far exceeds the industry standard, bringing the level of cockpit intelligence to a new height.

In addition, it can smoothly run various mainstream large language models (LLMs) and AI mapping functions in the car, providing users with advanced AI safety and entertainment applications such as 3D graphical interface based on in-car voice assistant, multi-screen interaction and display technology, and driving alert monitoring.

In terms of ecosystem development, MediaTek provides the NeuroPilot generative AI developer kit, which includes a large model center that supports the world's mainstream large language models.

The new tool chain incorporates NeuroPilot Fusion and LoRA technology (low-rank adaptive fusion), which can quickly complete model optimization and conversion.

The complete tool chain and documentation support enable developers to develop and deploy large models on the edge more easily.

It is worth mentioning that the Dimensity automotive platform shortens the design cycle of the original cockpit chip from the traditional 18-24 months to one year, keeping pace with the pace of new product launches by automobile manufacturers.

In addition, the underlying SDK and other resources are open to module manufacturers, allowing Tier 1 to quickly debug interfaces, thereby significantly shortening the development cycle and reducing costs to ensure product competitiveness.

Domestic smart cockpits are launched simultaneously, all with AI large models

SenseTime announced on April 23 the launch of the newly upgraded “SenseNova 5.0” large model.

This model combines the capabilities of multimodal large models, large language models, and literary graph models. Through SenseTime Jueying's innovative design, it has built a series of large-model cockpit products with panoramic perception, active care, and creativity, with a multimodal scene brain as the core.

At the Beijing Auto Show on April 25, CoreDrive Technology released a new generation of central processing units and domain controllers for automotive chips.

The new generation of X9CC is equipped with a high-performance NPU unit that supports local and cloud hybrid deployment of large models.

In the future, Xinchi will also launch the AI ​​cockpit processor X10, which will more efficiently support the Transformer architecture and realize pure end-side deployment of large models, thereby bringing users a safer, more efficient and personalized AI cockpit experience.

Huawei held a smart car solution launch conference on April 24, launching a new smart car solution brand with smart driving as its core - Huawei Qiankun.

The brand covers ten new products in four major areas: Qiankun Intelligent Driving, Qiankun Vehicle Control, Qiankun Car Cloud and Hongmeng Cockpit.

Among them, the new generation of Hongmeng cockpit is equipped with the Qianwu Engine large model, which is based on technologies such as Huawei Cloud Pangu large model, MindSpore computing framework and Ascend AI basic hardware platform.

Qianwu's visual function can achieve millimeter-level precise perception in the cabin and adaptive adjustment of the driver's seat. It also supports skeleton-level human perception in the entire cabin and multi-modal fusion vehicle control, bringing users an unprecedented in-car somatosensory interactive experience.

iFlytek has launched a new "Flying Fish Scenario Intelligent Cockpit System". The system is closely integrated with various in-car usage scenarios such as driving, communication, and entertainment by deeply integrating the core functions of large-scale model induction, reasoning, context understanding, and complex content generation.

Through the integrated use of voice and visual interaction, the system brings users a more natural, freer and more intelligent in-car experience.

Industry prospects and development trends

As an important force in the global automotive market, China's smart cockpit market development prospects have attracted much attention.

According to the forecast of Chuangye Research Center based on data from ICV Tank consulting firm and Visteon, the global smart cockpit industry market size will reach US$53.9 billion in 2022.

China, as a leader among them, also achieved a smart cockpit market size of 73.9 billion yuan that year.

Looking ahead, with the continuous advancement of technology and the continued expansion of the market, it is estimated that by 2025, the scale of China's smart cockpit market will exceed the 100 billion yuan mark, reaching 103 billion yuan, with an annual compound growth rate of 11.70%.

This growth trend shows the huge potential and broad prospects of China's smart cockpit market.

In the era of traditional fuel vehicles, the electronic control architecture of the car cockpit adopted a distributed electronic and electrical (E/E) architecture, that is, multiple independent ECUs controlled different functions.

With the development of the smart car industry, the E/E architecture of smart cockpits is evolving from distributed to domain-based, using domain controllers to control multiple system functions, reduce costs, and increase connectivity. In the future, the centralized domain controller E/E architecture will become the mainstream solution.

The smart cockpit domain and smart driving domain are key areas for the intelligence of new energy vehicles, and different chip platforms are usually responsible for providing computing power and control.