NIO reorganizes the intelligent driving R & D department？ The official has not responded yet. Experts： In order to adapt to the development trend of large models

On June 19, according to LatePost, NIO’s intelligent driving R & D department recently completed the structural adjustment. Previously, NIO’s intelligent driving R & D department was divided into perception, regulation and control, and integration. After the adjustment, the perception and regulation teams were merged into a large model team, and the integration team was reorganized into a delivery team. The merged large model team was overseen by Peng Chao, the former head of NIO’s perception team.

Regarding the authenticity of the above news, the Daily Economic News reporter immediately checked with the NIO, but no reply was received as of press time.

On the evening of June 20, an unnamed person familiar with the matter told reporters that NIO did make structural adjustments to its intelligent driving R & D team. However, the other party did not disclose the specific details of the adjustment.

Zhu Xichan, a professor at the School of Automotive Science at Tongji University and director of the Institute of Automotive Safety Technology, said in an interview with reporters that NIO’s move is more to adapt to the development trend of large models. "Tesla’s FSD V12 version has shown excellent autonomous driving capabilities end-to-end, and the autonomous driving AI technology route has been established." Zhu Xichan believes that after the automotive industry enters the second half of intelligence, the competition will mainly focus on autonomous driving technology.

Architecture adjustment Want to explore end-to-end large models?

According to public data, there are currently about 1,500 people in the NIO smart driving team. "With the launch of the big model, the autonomous driving technology has been transformed from the algorithm-based model of the past to the data-driven model, that is, the end-to-end big model." Ji Xuehong, a professor and director of the Automotive Industry Innovation Research Center at North China University of Technology, said that the integration of NIO’s sensing and regulation teams is the need for the organizational structure to serve the development of enterprise technology.

Image source: Photo by Zhang Jian, a reporter (data map)

It is reported that the restructured autonomous driving R & D department is still in charge of Ren Shaoqing, vice president of NIO’s intelligent driving R & D. After the adjustment of the intelligent driving R & D department, Ren Shaoqing once conveyed to the team: to give up the traditional paradigm of "perception-decision-regulation" that the industry has used for many years. This also means that NIO will more clearly explore the use of end-to-end large models to achieve high-level intelligent driving.

Zhu Xichan told reporters that NIO’s original modular autonomous driving algorithm team structure was divided into "perception", "regulation and control" and "evaluation", but after reaching the end-to-end large model, the division of departments is unreasonable. "In order to adapt to the new algorithm structure, the administrative structure of [NIO intelligent driving] can be optimized to break down departmental barriers and achieve more efficient development," Zhu Xichan analyzed.

Chen Guoping, executive director of Haosu Capital, told reporters that from the current point of view, the intelligent driving development program has been fully transformed into end-to-end autonomous driving, which has become an option for car companies. It turns out that perception, regulation, and decision execution are the R & D paths of traditional autonomous driving. The neural networks algorithm is used to form this simulated human driving decision through the accumulation of training data.

"With the comprehensive change of Tesla’s FSD driving decision model, the end-to-end large model output method skips the original traditional perception-decision-execution path, and calls the mainstream decision to drive the driving behavior in full accordance with the empirical algorithm, reducing the load on the perceived computing power technical parameters. The iteration speed will be faster and faster, and the cost of the smart driving solution will be lower and lower." Chen Guoping said that the new car manufacturing forces choose the end-to-end large model to merge and revoke the original organizational structure corresponding to the technology research and development path.

In an interview with reporters, He Xiaopeng, chairperson and CEO of XPeng Motors, said that in the previous smart driving scheme, technically speaking, the car was handled separately in terms of perception, positioning, planning, and control, and each link was not related. Therefore, when the vehicle encounters some scenarios, it will hesitate because of the rules written by humans. After the end-to-end large model gets on the car, the vehicle can directly output the control of the vehicle through the input of the sensor, and the smart driving will be more "human-like".

In He Xiaopeng’s view, the end-to-end smart driving model is adopted, which can be iterated every two days, and the smart driving capacity enhancement is 30 times in the next 18 months. The end-to-end big model means that automatic assisted driving will shift to fully autonomous driving.

Domestic car companies accelerate the deployment of autonomous driving

Speculation about the reasons behind the restructuring of NIO’s intelligent driving R & D department is mixed. Some people believe that NIO’s move is affected by the approaching time of Tesla’s FSD entry into China. However, Yang Weibin, an expert on new energy vehicles and batteries, told reporters that the adjustment of NIO’s intelligent driving R & D team is not directly related to Tesla’s FSD entry into China. "Because NIO and Tesla use two different intelligent driving technology routes, each has its own advantages and disadvantages, and it is not possible to completely judge which solution is better at present," Yang Weibin said.

Zhu Xichan also said that the reorganization of NIO’s intelligent driving R & D department has nothing to do with whether Tesla FSD enters China or not, but is mainly affected by the end-to-end model technology of Tesla FSD V12 version. "As the world’s leading company in autonomous driving technology, the gradual landing of Tesla FSD has pointed out the development direction and technical route of smart cars for the industry." Zhu Xichan said that with the improvement of the technical capabilities of the end-to-end model of Tesla FSD V12 version, even if it does not enter China, the autonomous driving technology of domestic car companies such as NIO will shift to the direction of large models.

In March this year, Tesla began to push the end-to-end FSD V12 to North American users on a large scale. Recently, it was reported that the Shanghai Autonomous Driving Demonstration Zone has issued a road test license to Tesla, and the FSD may be testing. In response, the reporter checked with Tesla China on June 21, but no response was received as of press time.

Ji Xuehong also said that Tesla is currently the best company in autonomous driving technology. Whether it enters China or not, it will guide more and more car companies to make more investment in autonomous driving technology. "As a’catfish ‘in the field of autonomous driving, Tesla will accelerate the layout and force of car companies in the field of autonomous driving." Ji Xuehong explained that in addition to NIO, XPeng Motors, Li Auto, Huawei and other car companies, more and more traditional car companies are also increasing the layout of autonomous driving technology direction.

On June 20, Zhang Xinghai, chairperson (founder) of Cyrus Group, announced that Cyrus will firmly pursue the development of smart-electric integration under software-defined vehicles, and strive to explore solutions for intelligent driving and intelligent safety. "At present, the AITO cars produced by Cyrus have driving auxiliary features, of which two-thirds have high-order intelligent driving functions, and the auxiliary driving functions account for one-third." Zhang Xinghai said.

In Yang Weibin’s view, car companies are doing theory in the "first half" and practice in the "second half". "Car companies will take advantage of the opportunity of large-scale demonstration of domestic smart driving technology, expand the number of cities where smart driving is applied, and continue to iterate and optimize and upgrade." Yang Weibin said that there is still a long way to go to fully realize autonomous driving.

Daily economic news