Currently, the intelligent cockpit is gradually becoming the third space of mobility. It is not only a hub connecting work and life, but also an information distribution center that integrates comfort and intelligence. On April 25, 2025, at the Automotive Consumption Trend Insight Conference, Duan Qi, the head of the Intelligent Cockpit Department of BAIC Group's Intelligent Connectivity Center, pointed out that the development of intelligent cockpits should not be limited to functional stacking, but should alleviate users' anxiety while driving through better design. With the development of computing power and intelligence, intelligent cockpits are evolving towards AI cockpits, which can automatically match and execute user needs, bringing drivers a more convenient and personalized experience.

In response to the challenges posed by the development of intelligent cockpits, BAIC Group has established the "Hundred Model Huichuang" central large model platform. This platform integrates vertical models within the industry to achieve atomic scheduling of vehicle capabilities and connect with external ecosystems, greatly enriching the functionality and experience of intelligent cockpits. Users can enjoy the full process service from travel planning to destination consumption guidance through voice commands.

BAIC Group: New Trends in the Development of Next Generation Intelligent Spaces

Quan Qi | Director of Intelligent Cockpit Department of BAIC Group Intelligent Network Connection Center

The following is a summary of the speech content:

Currently, intelligent cockpits are regarded as the third space of mobility, serving as hubs and bridges connecting workplaces, homes, and the next station. In the process of moving, people can adjust their mood and living conditions in this space. In the development process of intelligent cockpit, with the gradual increase of functions, there have been cases of hardware increase, and many physical switches have been transformed into virtual switches, which has actually brought pressure to car owners to a certain extent. We believe that developing an intelligent cockpit is not simply about stacking various functions on users. Users should focus their main energy on driving behavior during the driving process, and should not increase their pressure due to the addition of functions. Instead, they should alleviate their anxiety while driving through better design.

Breakthrough in the concept of automobiles

Smart cockpit and smart home are becoming increasingly similar, both consisting of overall layout and internal intelligent hardware. Looking back at the development history of automobiles, they were initially born as tools, enabling the movement of objects or people from point A to point B. In the process of development, cars have taken on the form of carports and enclosed spaces, effectively achieving changes in the environment within small spaces.

BAIC Group: New Trends in the Development of Next Generation Intelligent Spaces

Image source: Speaker materials

The atmosphere lights, air conditioning and other facilities equipped in the intelligent cockpit can create a more comfortable environmental atmosphere. In addition, the intelligent cockpit also has energy transfer function, such as the square box pure electric off-road vehicle in new energy electric vehicles, which can output stable energy to the outside when camping in remote areas without electricity.

With the development of computing power and intelligence, the intelligent cockpit is still becoming a mobile information distribution center. We expect it to serve as an information hub, not just for transmitting information, but to assist users in handling external affairs more effectively while driving.

With the development of cockpits, they initially evolved from functional cockpits to intelligent cockpits. The biggest feature is that the functional cockpit only has a single point-to-point function, while the intelligent cockpit, with the development of electronic and electrical architecture, no longer relies on a single attribute or configuration for function realization, but is formed by the combination of several small functions to form a large function. For example, it can achieve nap mode, movie watching mode, etc. Users can activate various functions on the car through simple operations. The next development of intelligent cockpit, namely AI cockpit, we believe that at this stage, there is no need for users to manually operate it. Users only need to say their requirements, and the cockpit can automatically execute it. This is the original intention we hope to achieve in the era of intelligent cockpit.

Previously, the industry has generated numerous scene combinations through the integration of various scenarios and the linkage of functions. Under these scene combinations, the number of scenes continues to increase, and the possibility of their arrangement and combination is almost astronomical. Initially, different scenarios were pre-set by car manufacturers, but due to the diversity of user needs, this approach posed difficulties. Later, in order to meet the personalized needs of users, manufacturers opened up functions to users, allowing them to define them independently, thus giving rise to programmable cockpits.

However, from user purchase data, programmable cockpit functions are powerful, but in actual use, the usage rate of car owners is sometimes not high. Most car owners only set a few functions in the early stages of purchasing a car, and rarely trigger them again when encountering other unsuitable scenarios, only as an initial trial. Nevertheless, these features can still match users' daily habits well. Therefore, we need a more flexible way to match users with corresponding scenarios in different situations.

The scenarios that product managers have painstakingly imagined in the past are difficult to meet the various possibilities that users face every day. With the development of AI big models, this seemingly impossible task has become possible. With the advancement of intelligent cockpit AI, intelligent cockpit should not only be a porter of functions and information, but should also provide services to users at the appropriate time and in the appropriate way. That is, to form the functions that an intelligent cockpit should have through various arrangements and combinations, as well as corresponding understanding and cognition in different times and spaces.

Thoughts on the Application of BAIC Model Technology

At present, the development of AI has been widely integrated into various industries and is developing rapidly. Based on this, we built a central large-scale model platform called "Baimo Huichuang" in 2024. The cost of training a single model for automobile manufacturers is enormous, especially for training basic models, and with the rapid development and iteration of the automotive industry, this cost pressure has become more prominent. Therefore, we build a central large model platform that takes control of the invariant parts while leaving enough flexibility for the mutable parts. On this platform, vertical models from various fields can be flexibly accessed and replaced. Simply focus on developing your own intelligent agent, while other capabilities and basic functions can come from various vertical models within the industry, and can be flexibly replaced. You can also choose the best performing product in a certain field within the industry.

BAIC Group: New Trends in the Development of Next Generation Intelligent Spaces

Image source: Speaker materials

Based on the overall architecture of "Baimo Huichuang", on the one hand, we will integrate all atomization capabilities of the vehicle end and provide vehicle status and control permissions to the large model scheduling platform. At the same time, the large model scheduling platform can integrate various vertical models and external ecosystems in the cloud, including connecting the Internet of Things, mini programs, and payment processes. In this way, users can collaborate with all the ecosystems around them with just one sentence.

We believe that the integration of AI into intelligent cockpits will change the entire development logic and mode. The overall feature of this architecture is that it does not rely heavily on independent vertical classification capabilities, but rather introduces existing external vertical classification models to achieve functionality, integrating as many ecosystems and capabilities as possible into the platform to fully integrate and generate new features and experiences. With the change of interaction logic, complex human-computer interaction design will no longer be necessary, just like communication between people, which can still be vividly communicated without the help of screens or complex devices. We believe that once AI gets on board, it will bring about significant changes and ultimately achieve a return to simplicity.

Our overall idea is to connect users' abilities at different times and locations based on this platform. Previously, the team had multiple brainstorming sessions to explore which functions should be installed on the vehicle. We believe that the overall interactive experience of travel and intelligent cockpit should be focused on providing freshness and convenience, and relevant functions should be created.

Chatting function is one of the fundamental abilities. After users have access to the large model, they will ask various questions. This feature is particularly useful when taking children on trips, as it can answer all kinds of strange and curious knowledge questions raised by children. In addition, in the past, obtaining wallpapers required purchasing copyrighted images from relevant companies and cropping them. Now, with the ability to generate images, although not a mandatory feature, it can reduce initial development investment and increase user playability.

We also carried out ecological integration. When we navigate to a certain place, instead of cold alternative addresses, we can see the Tiktok video of the destination for better planning and selection. After the entire ecosystem is established, the content that can be connected becomes more abundant. The purpose of travel is not only to park at the destination, but also to plan subsequent activities after getting off the car, such as whether to park the car at a slightly farther charging station to charge the vehicle while heading to the dining area, saving time.

We believe that the role of the big model in getting on the car is not simply to create gimmicks, but to help users achieve a coherent process from pre travel planning, travel advice, to destination time utilization and surrounding consumption guidance after travel, creating a product that better understands their own needs than users. In the end, we will build a more concise human-computer interaction framework, transforming the functions that originally required users to manually click into user directed vehicle execution.

Since the first quarter of last year, this platform has been gradually installed on products such as Arctic Fox Koala and Arctic Fox Alpha T5. As of now, the total number of adjustments has reached 51 million, covering almost all car owners. With the continuous improvement of functions, we expect to provide more diverse AI services for more car owners.

During the use of "Baimo Huichuang", users only need to state their requirements to combine the corresponding functions on the car. For example, when a user says "It's raining now, I need...", due to the fact that some functions inside the car may be hidden deep or unknown to the user, based on the ability of programmable cockpit and combined with the large model, the user only needs to put forward requirements to the large model, and the system can automatically combine the required functions. Previously, tedious clicking and rotating operations, as well as pre-set function combinations that were difficult to meet various daily situations, needed to be performed in a programmable cockpit. Now, users can complete them with just one sentence.

On February 8th of this year, we integrated DeepSeek as a feature of the "Baimo Huichuang" central platform for deep reasoning. We believe that the first step is to have a unified central platform to achieve functional connectivity between various models and their responses; Secondly, by continuously adjusting the combination of access models, the overall product capability can be further enhanced.

Prospects for the Application of Large Models

With the advent of the era of big models, we believe that there will be four directions of reconstruction. In terms of software, human-computer interaction will undergo changes, software architecture will become more atomic, and the precision of large-scale model scheduling will continue to improve; In terms of hardware, hardware design will no longer only meet current needs, but also reserve sufficient bandwidth to cope with unknown scenarios that may arise in the future.

As a traditional car company, we will increase more data integration and touchpoints in terms of data and ecology, establish user long short-term memory databases and vector database mechanisms, enrich the recording of user usage habits in the large model platform, and provide more personalized recommendations.

In terms of cooperation, we are no longer just working with traditional Tier 1 or Tier 2 partners. Given the strong backward consumer nature of automotive products, we should directly communicate and cooperate with platforms that provide post travel services to expand our circle of friends.

Finally, we believe that the impact of the big model industry on intelligent cockpits is positive. Despite encountering many difficulties and challenges during the development process, it is difficult to find a development path without actively engaging in practice. As with the development of emerging technologies in the past, there may be initial pains, but we believe that AI big models will inject new vitality and momentum into the entire automotive industry.

(The above content is from the keynote speech on "New Trends in the Development of Next Generation Intelligent Space" delivered by Mr. Qi Qi, the head of the Intelligent Cockpit Department of BAIC Group's Intelligent Connected Center, at the Automotive Consumption Trend Insight Conference on April 25, 2025.)