Autonomous driving and AI large-scale models "fall in love and kill each other": the large-scale model reshaping algorithm accelerates the arrival of the era of L3 autonomous driving

Original source: Financial Association

Image source: Generated by Unbounded AI

Financial Associated Press, July 9th (Editor Liu Yue) Speaking of AI technology, autonomous driving and large-scale models are two important issues. Recently, autonomous driving, which has been silent for a while, has become popular again. The Ministry of Industry and Information Technology will support the commercial application of L3 and higher levels of autonomous driving functions, Musk said on Thursday that full autonomous driving is expected to be achieved by the end of this year. Tesla FSD and Xiaopeng G6 have also caused heated discussions in turn. In the secondary market, the smart driving concept stock ** Kotei Information has a daily limit of 20cm on Tuesday **, Zhejiang Shibao achieved five consecutive boards on Thursday, and the stock price has accumulated the largest increase of 119.73% since the low point on June 8. The stock price of Desay SV has risen by more than 90% during the year.

The intelligent driving track gathers all kinds of players, including Tesla, Xiaopeng Motors, Weilai and other new car-making forces, BYD, SAIC Group, GAC Group, Changan Automobile, Dongfeng Motor and other traditional overall players Depots, Huawei, Baidu, Apple, Xiaomi, Alibaba and other technology companies, as well as momenta, Didi and other start-up companies.

The research report of Soochow Securities sorts out the top-level autonomous driving architecture and progress of car companies such as Great Wall Motors, BYD, Changan Automobile and Celes**, and points out that 2023 H2 and 2024 are the window period for car companies to accelerate the launch of intelligent new cars , among which the Urban NOA (L3-like autonomous driving) function has become the key capability to distinguish the intelligence of car companies.

Large model reshaping autonomous driving algorithm Technology giants provide AI large model development services for OEMs

Musk has set the target time for fully automatic driving many times before, but he has not achieved it. Why is he so confident about "realizing this year" now? Perhaps it is the technological development in this year's AI wave that has added a "fire". The CEO of Alibaba Group once said: "In the era of AI, all products are worth re-upgrading with large models." Naturally, autonomous driving products are no exception. The research report released by Kaiyuan Securities Ren Lang on July 2 pointed out that the latest large-scale model reshapes the automatic driving algorithm, and driven by the large-scale model, L3-level automatic driving is accelerating. **According to data from iiMedia Consulting, the scale of my country's driverless car industry is expected to grow from 3.05 billion yuan in 2015 to 26.76 billion yuan in 2025, with a CAGR of 24.3%.

Kaiyuan Securities pointed out that there is a clear trend in the application of large models in autonomous driving. In the cloud, the capacity advantage brought by the increase in the number of large model parameters can be used for automatic driving data automatic labeling, data mining, training small models through distillation, etc.; On the car side, large models are available In terms of merging small models used for different detection tasks, it saves the inference time required for the calculation of the car and increases the safety of automatic driving. At present, new energy vehicle manufacturers including Xiaopeng and Tesla are accelerating the implementation of driving assistance functions on urban roads**. With the accelerated empowerment of AI large models, the "singularity" of autonomous driving is always expected to accelerate.

A research report released by Orient Securities Kuai Jian et al. on June 27 pointed out that according to data from the Ministry of Industry and Information Technology, the penetration rate of L2-level assisted driving in 22 years is 34%, and the penetration rate of L3 autonomous driving in 30 years will reach 70%. AI is the key to autonomous driving modular systems and end-to-end systems: mainstream modular autonomous driving systems can be divided into three layers: perception, decision-making, and execution. AI algorithms are the core of the perception layer and decision-making layer, while end-to-end In the system, input data to output control is realized only through a large AI model. The development of AI large-scale model technology allows autonomous driving technology to remove its dependence on high-precision maps by improving perception capabilities.

Technology giants can provide computing power support and AI large-scale model development services for autopilot OEMs. Changan, Jidu, Geely, Lantu, Hongqi, Great Wall, Dongfeng Nissan, Leapmotor and many other car companies have announced that they will access Wenxinyiyan. Changan Yida became the first model to be equipped with Wenxin Yiyan, and it will be installed on the new car through software upgrades. Huawei announced in May that the AITO M9 will be equipped with a large AI model, and Xiaoyi's smart assistant will have a better in-vehicle AI experience. HKUST Xunfei "Spark Cognition" large-scale model also has products related to the automotive field, which can realize cross-business and cross-scenario human-vehicle free communication in the car.

Large models "eat" autonomous driving: DriveGPT is unrealistic to reduce costs and increase efficiency, Baidu and other major manufacturers shrink their autonomous driving business

Under the heat, the self-driving industry is also chilling. According to media reports, young people working on autonomous driving were laid off three times in two years. According to market analysis, L4 has been constrained by the "impossible triangle" of security, data, and cost in the past two years. At the same time, software delays such as algorithms are only piled up with hardware, which has caused leading companies to fall into a double dilemma of technology and commercialization. Bureau **. Some analysts also pointed out that new energy vehicles have crossed the gap and been accepted by the market. But "autonomous driving" is just a carnival for technology enthusiasts and has not yet been accepted by the market.

In addition, some people in the industry said that DriveGPT is very unrealistic. Even if large companies invest in research and development, it will be difficult to see results within 5-10 years. If the automatic driving system is to be used on a large model, it will cost at least US$50,000. As the large model becomes larger, the cost may increase further. Being unable to get in the car has become the primary problem that plagues the commercialization of large-scale autonomous driving models. Orient Securities also pointed out that the application of AI large models is still short, and the model architecture in the field of autonomous driving is still being explored.

In big factories, big models even "eat" automatic driving. Baidu, which used to be the most steadfast in autonomous driving, also began to adjust its business. According to Times Finance reports, recently, Baidu adjusted the organizational structure of the Intelligent Driving Business Group (IDG), and its Intelligent Transportation Business Unit (ACE) was assigned to the Intelligent Cloud Business Group (ACG). After the adjustment, IDG's business segments have been adjusted from the original three major segments of autonomous driving, smart cars, and smart transportation to only the first two.

It is reported that this is not the first time that Baidu’s autonomous driving-related business has changed this year. In January this year, Baidu IDG was revealed to be laying off personnel. In early June, Guo Yang, chief product architect of IDG, resigned. More importantly, the phenomenon of business shrinkage in the autonomous driving field is not limited to Baidu alone**. This year, Alibaba split the DAMO Institute and merged the self-driving laboratory into Cainiao; the self-driving truck company Optima Zhika, which was diverted from Pony.ai, went bankrupt after 19 months of operation; Tencent, which announced its entry into autonomous driving in 2018, has long been invisible to the public. Overseas, Uber sells its autonomous driving team, Amazon shuts down the unmanned delivery vehicle project, and Tucson Future, known as the "first autonomous driving stock", will carry out a new round of large-scale development in May layoffs.

Analysts said that under the background of Internet companies generally emphasizing cost reduction and efficiency increase, when the new AI large-scale model comes, the former core business of autonomous driving has to be sidelined. After a large number of major manufacturers recognized the difficulty of autonomous driving, the AI model became a stimulant for the major technology companies to get rid of its shadow. Financial Associated Press, July 9th (Editor Liu Yue) Speaking of AI technology, autonomous driving and large-scale models are two important issues. Recently, autonomous driving, which has been silent for a while, has become popular again. The Ministry of Industry and Information Technology will support the commercial application of L3 and higher levels of autonomous driving functions, Musk said on Thursday that full autonomous driving is expected to be achieved by the end of this year. Tesla FSD and Xiaopeng G6 have also caused heated discussions in turn. In the secondary market, the smart driving concept stock ** Kotei Information has a daily limit of 20cm on Tuesday **, Zhejiang Shibao achieved five consecutive boards on Thursday, and the stock price has accumulated the largest increase of 119.73% since the low point on June 8. The stock price of Desay SV has risen by more than 90% during the year.

The intelligent driving track gathers all kinds of players, including Tesla, Xiaopeng Motors, Weilai and other new car-making forces, BYD, SAIC Group, GAC Group, Changan Automobile, Dongfeng Motor and other traditional overall players Depots, Huawei, Baidu, Apple, Xiaomi, Alibaba and other technology companies, as well as momenta, Didi and other start-up companies.

The research report of Soochow Securities sorts out the top-level autonomous driving architecture and progress of car companies such as Great Wall Motors, BYD, Changan Automobile and Celes**, and points out that 2023 H2 and 2024 are the window period for car companies to accelerate the launch of intelligent new cars , among which the Urban NOA (L3-like autonomous driving) function has become the key capability to distinguish the intelligence of car companies.

Large model reshaping autonomous driving algorithm Technology giants provide AI large model development services for OEMs

Musk has set the target time for fully automatic driving many times before, but he has not achieved it. Why is he so confident about "realizing this year" now? Perhaps it is the technological development in this year's AI wave that has added a "fire". The CEO of Alibaba Group once said: "In the era of AI, all products are worth re-upgrading with large models." Naturally, autonomous driving products are no exception. The research report released by Kaiyuan Securities Ren Lang on July 2 pointed out that the latest large-scale model reshapes the automatic driving algorithm, and driven by the large-scale model, L3-level automatic driving is accelerating. **According to the data from iiMedia Consulting, the scale of my country's driverless car industry is expected to grow from 3.05 billion yuan in 2015 to 26.76 billion yuan in 2025, with a CAGR of 24.3%.

Kaiyuan Securities pointed out that there is a clear trend in the application of large models in autonomous driving. In the cloud, the capacity advantage brought by the increase in the number of large model parameters can be used for automatic driving data automatic labeling, data mining, training small models through distillation, etc.; On the car side, large models are available In terms of merging small models used for different detection tasks, it saves the inference time required for the calculation of the car and increases the safety of automatic driving. At present, new energy vehicle manufacturers including Xiaopeng and Tesla are accelerating the implementation of driving assistance functions on urban roads**. With the accelerated empowerment of AI large models, the "singularity" of autonomous driving is always expected to accelerate.

A research report released by Orient Securities Kuai Jian et al. on June 27 pointed out that according to data from the Ministry of Industry and Information Technology, the penetration rate of L2-level assisted driving in 22 years is 34%, and the penetration rate of L3 autonomous driving in 30 years will reach 70%. AI is the key to autonomous driving modular systems and end-to-end systems: mainstream modular autonomous driving systems can be divided into three layers: perception, decision-making, and execution. AI algorithms are the core of the perception layer and decision-making layer, while end-to-end In the system, input data to output control is realized only through a large AI model. The development of AI large-scale model technology allows autonomous driving technology to remove its dependence on high-precision maps by improving perception capabilities.

Technology giants can provide computing power support and AI large-scale model development services for autopilot OEMs. Changan, Jidu, Geely, Lantu, Hongqi, Great Wall, Dongfeng Nissan, Leapmotor and many other car companies have announced that they will access Wenxinyiyan. Changan Yida became the first model to be equipped with Wenxin Yiyan, and it will be installed on the new car through software upgrades. Huawei announced in May that the AITO M9 will be equipped with a large AI model, and Xiaoyi's smart assistant will have a better in-vehicle AI experience. HKUST Xunfei "Spark Cognition" large-scale model also has products related to the automotive field, which can realize cross-business and cross-scenario human-vehicle free communication in the car.

Large models "eat" autonomous driving: DriveGPT is unrealistic to reduce costs and increase efficiency, Baidu and other major manufacturers shrink their autonomous driving business

Under the heat, the self-driving industry is also chilling. According to media reports, young people working on autonomous driving were laid off three times in two years. According to market analysis, L4 has been constrained by the "impossible triangle" of security, data, and cost in the past two years. At the same time, software delays such as algorithms are only piled up with hardware, which has caused leading companies to fall into a double dilemma of technology and commercialization. Bureau **. Some analysts also pointed out that new energy vehicles have crossed the gap and been accepted by the market. But "autonomous driving" is just a carnival for technology enthusiasts and has not yet been accepted by the market.

In addition, some people in the industry said that DriveGPT is very unrealistic. Even if large companies invest in research and development, it will be difficult to see results within 5-10 years. If the automatic driving system is to be used on a large model, the cost will increase by at least US$50,000. As the large model becomes larger, the cost may increase further. Being unable to get in the car has become the primary problem that plagues the commercialization of large-scale autonomous driving models. Orient Securities also pointed out that the application of AI large models is still short, and the model architecture in the field of autonomous driving is still being explored.

In big factories, big models even "eat" automatic driving. Baidu, which used to be the most steadfast in autonomous driving, also began to adjust its business. According to Times Finance reports, recently, Baidu adjusted the organizational structure of the Intelligent Driving Business Group (IDG), and its Intelligent Transportation Business Unit (ACE) was assigned to the Intelligent Cloud Business Group (ACG). After the adjustment, IDG's business segments have been adjusted from the original three major segments of autonomous driving, smart cars, and smart transportation to only the first two.

It is reported that this is not the first time that Baidu’s autonomous driving-related business has changed this year. In January this year, Baidu IDG was revealed to be laying off personnel. In early June, Guo Yang, chief product architect of IDG, resigned. More importantly, the phenomenon of business shrinkage in the autonomous driving field is not limited to Baidu alone**. This year, Alibaba split the Bodhidharma Institute and merged the self-driving laboratory into Cainiao; the self-driving truck company Optima Zhika, which was diverted from Pony.ai, went bankrupt after 19 months of operation; Tencent, which announced its entry into autonomous driving in 2018, has long been invisible to the public. Overseas, Uber sells its autonomous driving team, Amazon shuts down the unmanned delivery vehicle project, and Tucson Future, known as the "first autonomous driving stock", will conduct a new round of large-scale development in May layoffs.

Analysts said that under the background of Internet companies generally emphasizing cost reduction and efficiency increase, when the new AI large-scale model comes, the former core business of autonomous driving has to be sidelined. After a large number of major manufacturers recognized the difficulty of autonomous driving, the AI model became a stimulant for the major technology companies to get rid of its shadow.