On April 18, 2024, the second Automotive artificial Intelligence Conference 2024, hosted by Galaxy Automotive, was successfully concluded in Shanghai.
Since the beginning of this year, topics such as AI agent, big models and so on have attracted wide attention.
Focusing on the automotive industry, related technologies such as AI and big models also bring more possibilities for automotive applications.
There is even an opinion in the industry that the end of car intelligence is AI.
So what’s the end of AI? Math? Algorithm? data? Or energy? As one of the official activities of the Shanghai Automotive Culture Festival, the second Automotive artificial Intelligence Conference 2024 invited representatives of 12 car companies and 12 technology companies to participate– AI meets cars: subversion and reshaping.
Participants shared and discussed hot topics such as large model technology, ChatGPT, LLM, computing algorithm, cloud native, end-to-end, automotive robot, AI smart driving, AI intelligence module and so on.
At the same time, the conference got Neousys accolade Technology, Zero Beam Technology, iFLYTEK, Spitchi, Weizhi Technology, Baidu Intelligent Cloud, Microsoft, Zhiji AI, Zhiyuan Electronics, Tongxing Intelligence, thinking Technology, Yi Pei Jie, Oriental Zhongke, Zhisheng Software, Ruisuang Weinar, Keloda, Fupi Ti, Youisai, Faben Information, Etechi, Zhijia Automobile, New Siwei, Shisheng Technology, billion-Mak Communications, Strong support from ecological partners such as Xinbang Holdings.
Smart Future: intelligent cockpit combined with the exploration and practice of AI technology, public data show that there are more than 19 language model R & D manufacturers in China, and 15 of them have approved the record of model products.
It is predicted that in 2023, the market size of China’s large language model has reached 13.
23 billion yuan, with a growth rate of 110%.
The market scale has increased rapidly, and the application scenarios have been continuously enriched.
Zhao Henry, chief engineer of intelligent cockpit of Beijing Automotive Research Institute Co., Ltd., said frankly: “AI for the improvement of efficiency, will be integrated into every corner of various industries.
” Zhao Henry | Chief Intelligent cockpit of Beijing Automotive Research Institute Co., Ltd. What is the impact of the Al model on smart cars? Henry Zhao believes that, first of all, the AI model will help smart cars better understand the traditional vehicle control intention.
secondly, based on the AI model, it will continue to improve driving safety, optimize user experience, and enhance cockpit comfort.
finally, as a collection of scenes and functions, the AI model will help the establishment of the “third living space”.
For the mainframe factory, computing power, data and application are the important support of the AI model.
At present, BAIC Research Institute has also carried out a lot of practice on the AI model to continuously expand the depth and breadth of the intelligent cockpit and enhance users’ emotional resonance and immersion.
Looking to the future, Zhao Henry said that the big model, as the core of the cockpit AI, is responsible for dealing with the interaction between users and the cockpit system, as well as intelligent decisions within the system.
Based on the development of the traditional cockpit around south (hardware) and north (software), multi-directional extension will be realized to build a more intelligent, efficient and seamless cockpit experience.
The exploration and practice of the big model, the application and landing of the big model in the automobile industry, two months after the release of ChatGPT, the number of users quickly exceeded 100 million, and the big model opened a new era.
More and more enterprises will quickly apply new large models in the next 3-5 years.
Zhang Fan, AI COO of Zhisu, said that the current large model has achieved the unity of data, algorithm, model and application, and one framework can solve all problems.
The direct result is that “AI has become more powerful and the cost of using AI has been reduced by a hundred times”, so the popularity of the AI model has become a consensus in the industry.
Zhang Fan | Zhisu AI COO, Zhisu AI was founded in 2019, formerly from the knowledge transfer of the computer Department of Tsinghua University.
In 2021, AI released tens of billions of models, and in 2022, it released hundreds of billions of models.
At the same time, it participated in the evaluation of the global mainstream model of Stanford.
AI has become the only selected model in Asia and is in the middle and upper reaches of the world.
According to Zhang Fan, Zhisu AI’s large model matrix fully covers text, code, images, and video.
In January 2024, Zhisu AI launched a new GLM-4, with an overall performance improvement of 60%, approaching GPT-4.
What practice will the big model have in the automobile industry? Zhang Fan believes that the first is the application of the intelligent cockpit, and the big model will help car companies create a smarter, more comfortable and easier to use third living space.
to achieve complex car control, car book Q & A, chat companionship, life services, multimodal interaction and other intelligent, scene-based applications.
Secondly, the big model will empower each marketing mode of the automobile.
Including customer label extraction, sales skills quality inspection, marketing skills assistance and sparring and so on.
The last is enabling after-sales service, including intelligent service, public opinion analysis and so on.
Next, the big model will cover the whole process of automobile R & D and manufacturing, and continue to improve the efficiency of automobile production.
As a professional conversational artificial intelligence platform enterprise in China, Spichi always takes “communicating everything and taking care of everything” as its mission to provide intelligent human-computer interaction software products, software and hardware integrated artificial intelligence products and conversational artificial intelligence technical services.
At present, the service covers smart cars, smart home appliances, consumer electronics and other areas of the Internet of things, as well as digital government and enterprise customers.
Zeng Chunhua, senior product director of the intelligent automobile division of Spitchi technology, said that with the development of intelligent vehicles, in order to meet the intelligent voice needs of human-computer interaction, Spitchi has set up a perfect core technology capability system, including full-link voice and language interaction technology, soft-hard integrated man-machine dialogue system construction capability, and large-scale dialogue system customization.
Related voice products have landed on many well-known models in the industry.
Zeng Chunhua | Senior Product Director of Smart car Division of Spitz Technology, based on conversational AI technology, Spitchi continues to upgrade the pedestal model to go deep into the on-board voice scene.
Zeng Chunhua introduced that the combination of DFM large model technology and full-link dialogue AI technology, deep ploughing vertical domain, pay attention to the landing of industry large model industry.
At the same time, the big model of Spice continues to shape the scene.
Red Hat OpenShift Al boosted the car company’s big model application scene, and Red Hat ranked among Top 20 AI cloud manufacturers on the 2024 CRN AI100 list.
What role does Red Hat play in the wave of AI models? According to Liu Yang, chief technical architect of Red Hat China: “just as we have done in the field of applications, combining ecosystem advantages, Red Hat provides a hybrid cloud platform for training, fine-tuning, testing, deployment and monitoring models, as well as complete lifecycle management.
” , Liu Yang | Chief of Red Hat ChinaTechnical architect, what are the challenges encountered by current users in implementing AI? Liu Yang shared from the aspects of low utilization of GPU, lack of distributed computing support, lack of unified platform, complex security and compliance requirements, difficulties in model monitoring and maintenance, and high requirements for technical support and training, and introduced the Red Hat AI/ML scheme in detail to address the above challenges.
The trend of intelligence in the automotive industry and AI computing platform, autopilot, intelligent cockpit and Internet service are the three key areas of AI-enabled smart vehicles.
Ma Guojian, director of Huawei’s self-driving AI computing platform, said frankly that the hottest thing is autopilot, which brings revolutionary experiences and changes.
At present, there are three trends in autopilot: first, data multiplication, in which AI data is the top priority.
second, the model converges to the third BEV+Transformer+Occ.
, and the frequency of iteration and training is getting faster and faster.
Tesla has been able to iterate once a day, Huawei iterates every 5 days, individual module iterations will be faster.
Ma Guojian | Director of Huawei’s self-driving AI computing platform, the big model is reengineering the whole business process of the automotive industry, and the need to reduce costs and increase efficiency is accelerating the commercial landing of AI.
In this context, Huawei Teng provides a full-stack independent, open, compatible, flexible deployment of smart drive cloud-based AI computing platform.
Ma Guojian gave a technical explanation around the “Teng Teng autopilot solution” and shared it in detail with specific landing cases.
From perception to the development and application of cognitive-autopilot perception technology, the perception system includes perception module, fusion module, location module and integrated system, which can be divided into perception 1.
0 and perception 2.0. She Feng, head of driving awareness at Geely Automotive, said that the 1.
0 stage is a traditional post-fusion method, in which each sensor is dispersed and then fused after sensing the target, which will cause division in actual mass production.
Stage 2.
0 evolved to pre-fusion BEV perception, which is equivalent to a God perspective, which can improve perceptual performance and accuracy, and can also be used as input for downstream users.
At present, there are two mainstream routes about intelligent driving perception system: “pure visual perception” and “multi-sensor fusion”.
She Feng frankly said that BEV pure visual perception has several advantages, including BEV perspective to make positioning more accurate, stable speed, while reducing the difficulty of regulation and debugging.
structural advantages, BEV pure visual architecture can integrate different sensors into the BEV architecture, improve development efficiency and optimize the allocation of manpower.
in addition, it can achieve high-precision maps, high fresh-keeping, high coverage.
But at the same time, visual perception is also faced with some technical challenges, such as pure visual BEV pre-fusion, model lightweight, de-precision mapping and so on.
In addition, she Feng said that in the perception process of autopilot, it is also faced with the challenges of virtual simulation technology-including data-driven scene automation construction, simulation platform cloud deployment, high concurrency testing, and physical-level multi-source sensor modeling.
From a forward-looking perspective, she Feng pointed out that on the one hand, it has become an industry consensus for BEV+Transformer to improve its perception and generalization ability.
on the other hand, BEV+Transformer has become the mainstream trend of autopilot algorithms.
Finally, in view of the development of the big model on autopilot, she Feng threw ten questions to the industry, with a view to causing more collisions of views.
Think deeply: how AI empowers smart cars.
In 2023, Chery Group served a total of 13 million customers, including 3.
35 million overseas users.
In 2023, Chery Group sold a total of 1.
881 million vehicles, of which Xingtuxing era, as the brand of the pyramid, has set up 300 stores nationwide by 2023, and is expected to reach 460 + in 2024.
According to Zhang Liyuan, director of Chery Automobile artificial Intelligence Technology Center and head of intelligent cockpit function development, Chery Group has made six achievements in the field of AI in 2023: first, cross-domain interaction.
Second, health monitoring.
Third, multi-round interaction.
, fourth, multilingual partner.
Fifth, minutes of the meeting.
, sixth, large model base.
Zhang Liyuan | Director of Chery Automobile artificial Intelligence Technology Center and person in charge of intelligent cockpit function development.
Around the above six points, Zhang Liyuan explained in detail with specific landing cases.
Looking to the future, Zhang Liyuan believes that, first of all, after the emergence of the big model, the application of AI in the automobile shows an obvious trend of de-app.
secondly, the advantages of the regulation AIGC will empower the automobile industry.
AIGC automotive applications will play an important role in automotive design, HMI, sales and marketing processes.
In the era of large model, the thinking and practice of the application of AI in automobile enterprises benefited from the accumulation of data and the development of computing power, and the big model came out of nowhere, triggering a new wave of AI.
Kuang Xu, head of the data algorithm and Model Department of Changan Automobile big data Center, said that as far as the automobile industry is concerned, AI+ has become the latest trend to replace Internet +.
Putting AI in the process of product definition can more accurately mine the needs of users, and at the same time, it can imagine the product form based on AI, which will be a new change to the definition of automobile products.
, Kuang Xu | head of data algorithm and Model Department of big data Center of Changan Automobile, as an intelligent mobile space, cars in the future will not only carry new technologies such as AI, chips and communications, but also have more interaction with users, become platform products, and innovate business models based on platform products.
What kind of AI capabilities do automobile enterprises need to have? Kuang Xu believes that it can be considered from two aspects: first, automobile companies must precipitate their own data assets.
Second, the construction of scene layout tools.
Chang’an Automobile hopes to use large models as the basis of the intelligent brain and turn small models into different tool chains, so that the brain can use tools effectively and attach equal importance to both large models and small models.
Based on the above thinking, Changan Automobile has carried out a series of AI application practices.
In the future, Changan Automobile will also build a unified AI infrastructure, support intelligent products and digital intelligence management, promote product and technological innovation, improve business efficiency, and lead the development of AI application and ecology in the transportation industry.
AI large model enables intelligent cockpit, when AI big model comes, enterprises can make full preparations from four aspects.
First of all, it is the preparation of the mind.
secondly, it is easy to do.
thirdly, it is necessary to find out what new experiences and changes the AI agent will bring.
and finally, when the big model within the enterprise comes, it can really achieve efficient development.
Centering on the above four points, the chief software development engineer of Hezhong New Energy Automobile Co., Ltd.Cai Yong shared his thoughts.
Cai Yong, Chief Software Development engineer of Hezhong New Energy Automobile Co., Ltd., which is about “what is the easiest thing?” Cai Yong said that voice in the cockpit is most easily empowered by large models, and voice has entered the daily life of users.
Currently, the user data of Nezha Voice shows that the intended usage is more than 800.
the average daily active rate is more than 80%.
and the average daily valid instruction is more than 10.
Therefore, Nezha voice focuses on natural, comprehensive and intelligent, which is to meet the personalized needs of users.
Next, the large model will help the voice interaction to a higher level, and the AI agent will bring new man-machine relationship.
In the exploration and application of Jianghuai Automobile in the large model, everything has its own law of development, and the same is true of the big model.
Xu Ruixue, a software development expert at Jianghuai Automobile Intelligent Network, admitted that in 2017, Google proposed the Transformer architecture, which is the first year of large model technology.
The development law of large model technology is as follows: from single-mode to multi-mode, from discrimination to generation, from special to general, and finally realize artificial general intelligence AGI in the ideal state.
Xu Ruixue | Jianghuai Automotive Intelligent Network Software Development expert, focusing on the automotive field, data-driven, large models link enterprises and vehicle users, covering R & D, production, marketing, operation, car purchase, car use, car maintenance, car repair, and after-sales lifecycle scenarios, providing simple and efficient knowledge digitization services.
Jianghuai Automobile has carried out practical applications around the large model of cockpit system, including natural interaction mode, efficient task processing, reliable AI description, interesting content creation and so on.
At the same time, the iHearing intelligent cockpit, the world’s first iFLYTEK spark cognitive model, has 19 bright functions, such as free chat and empathy dialogue, smart car experts, immersion scene service, intellectual and educational interactive assistant, and so on.
With regard to the large model of the intelligent driving system, Xu Ruixue said that the Jianghuai intelligent driving system design-perceives a network to plan a network, simplifies the perception and planning modules to a neural network, and can be continuously trained and updated in a data-driven way.
In addition, the data will be one of the key factors in the second half of the autopilot track, so Jianghuai is also laying out its own data closed-loop system.
With regard to the present situation and future of the intelligent cockpit enabled by AI, Great Wall Automobile has completed the intelligent industrial layout around the intelligent cockpit, intelligent driving and intelligent chassis.
It is worth mentioning that as a brand of intelligent technology for the whole vehicle of Great Wall, Coffee Intelligence can provide a global intelligent solution for the whole vehicle, and become the technical base to realize the strategic goal of “Green Wisdom play” of Great Wall Automobile.
According to Wu Xiugen, a senior technical expert at Great Wall Automobile AI, Coffee Intelligence adheres to the technical concept of “security, stability and integration of three intelligences” with user experience and demand as its traction, and will create a future travel partner that integrates the end of the cloud, the Internet of everything, can think, can judge and can grow.
At the same time, Great Wall Motor has realized the full-link self-research and development of vertical AIGC model to generate situational and interesting AI wallpaper for Great Wall mass production models, and empower the enterprise end.
Wu Xiugen | AI senior technical expert of Great Wall Automobile, about the technical layout of the next generation AI.
Wu Xiugen believes that the first is the deployment of multimodal algorithms.
the second is to focus on the application of large models, making full use of computing power and resources, while balancing the needs and capabilities of large models.
and the construction of the third AI Agent is the key to determine the ability to solve follow-up problems.
At present, the development of artificial intelligence has completed the transformation from “a large number of models” to “mass models”.
Liu Haifeng, chief engineer of software engineering at the General Research and Development Institute of Dongfeng Automobile Group Co., Ltd., said that the current large model shows three characteristics: letting a hundred flowers blossom and a hundred schools of thought contend.
open source research, closed source commercial use.
and single-mode to multi-mode.
At the same time, the large model can be applied to chat AI, video AI, drawing AI, programming AI and so on.
Liu Haifeng | Chief Software Engineering engineer of the General Research and Development Institute of Dongfeng Automobile Group Co., Ltd., at the same time, Liu Haifeng said that based on the basic capabilities of data platform, computing platform and cloud platform, aiming at people, cars and lives, the application fields of the automobile industry include general purpose, R & D process, product definition, product technology, manufacturing, after-sale, daily life, smart home, meta-universe and so on.
A total of 29 types of application scenarios.
Based on the large visual model, Dongfeng Automobile constructs an automatic tagging platform, which combines generative artificial intelligence and reinforcement learning technology to generate simulation scenes and data, realize automatic data classification, mining and labeling, and enable application scenarios such as self-driving and intelligent cockpit to facilitate the efficient iteration and delivery of AI models and software.
At this point, the two-day agenda of the General Assembly is all over.
As more and more automobile companies choose to access the AI model and redefine the “human-computer interaction” and “service ecology” of the automobile, the automobile industry will usher in the reconstruction of the industrial chain and the innovation of the business model.
Artificial intelligence will promote the deep integration of automobile and Internet, big data, artificial intelligence and other industries, and gradually form a new industrial ecosystem.
, return to the first electric network home page >.
