Li Auto Establishes 'End-to-End Autonomous Driving' Organization, Unveils Future Plans
TapTechNews July 18th news, according to the report of 36Kr Auto, Li Auto has recently established a physical organization of end-to-end autonomous driving, with a total of a little over 200 people, and some members of other teams in the company are also flexibly supporting the project.
It is reported that Li Auto's intelligent driving team is mainly divided into two large groups: algorithm R & D and mass production R & D. The team has about 800 people. The main R & D force of end-to-end is deployed in the algorithm R & D group, which is in charge of Jia Peng, the head of Li Auto's intelligent driving technology R & D.
The algorithm R & D group has subgroups such as perception algorithm, behavioral intelligence, and cognitive intelligence. Perception algorithm is led by Chang Li; Behavioral intelligence includes end-to-end architecture, end-to-end model, control model, etc., which is in charge of Yang Yi; Cognitive intelligence includes cognitive model, cloud model, etc., which is actually in charge of Zhan Kun.
The mass production R & D group is in charge of Wang Jiajia, including intelligent driving, intelligent parking, intelligent safety, etc. Both Jia Peng and Wang Jiajia report to Lang Xianpeng, the head of Li Auto's intelligent driving.
TapTechNews noticed that in June this year, Li Xiang, the chairman and CEO of Li Auto, disclosed at the 2024 China Auto Chongqing Forum that in the third quarter, an end-to-end + VLM (visual language model) supervised autonomous driving system trained by 3 million Clips (pre-trained model) will be launched to test users. From the end of this year to the beginning of next year, a supervised autonomous driving system trained by more than 10 million Clips will be launched.
During the Li Auto Intelligent Driving Summer Launch Event held on July 5th, Li Auto announced that the early bird plan of end-to-end + visual language model has officially started, claiming that it can make the car smarter and more like a human.
Li Auto claims that the advantages of the end-to-end model lie in two aspects: efficient transmission and efficient computing: end-to-end is an integrated model, and the information is transmitted inside the model, which has a higher upper limit, and the actions and decisions of the entire system that users can feel are more human-like; the integrated model can complete the reasoning in the GPU at one time, and the end-to-end latency is lower, and users can perceive that the eyes and hands are coordinated, and the vehicle's action response is timely.