The end-to-end model is involved in the smart driving circle Zhou Guang: Get on the car this year!

Letting AI drive like a human driver will be possible this year.
 
On March 17, at the China Electric Vehicle 100 Automobile New Quality Productivity Forum , Yuanrong Qixing CEO Zhou Guang said that the end-to-end intelligent driving model will create a "super AI driver" and open up general artificial intelligence technology in the physical world. new era. At present, the company has successfully adapted the end-to-end model to a mass-produced vehicle, which will be put into the consumer market this year.
 
Yuanrong Qixing is the first domestic artificial intelligence company that can successfully put an end-to-end model on the bus. In the current domestic consumer market, intelligent driving systems still use modular models, dividing perception, prediction, and planning into three independent models. The technology stack of each model is quite different, and the downstream planning model requires engineers to write a lot of code to formulate driving rules. Under the modular technical architecture, the transmission of information will be impaired, the maintenance of the system will be difficult, and it will not be able to deal with complex road conditions calmly. 
 

Zhou Guang, CEO of Yuanrong Qixing
 

The end-to-end model is completely different. This model integrates the three models of perception, prediction, and planning. It does not require programmers to write lengthy codes to formulate rules. Instead, it uses massive data to train the system and empowers the machine to learn and think autonomously. and analytical skills. The end-to-end model will not suffer from loss of information transmission, can better handle complex driving tasks, and solves all the "pain points" of the modular model.
 
"The transmission of intelligent driving information should not be about drawing and guessing, but the most direct response to the information seen like humans. The modular model decomposes a complex task into multiple sub-tasks and solves them one by one. This way Problem-solving methods are tortuous and error-prone. The end-to-end model is based on the root of the problem and directly seeks the optimal solution," Zhou Guang said.
 

Yuanrong Qixing's end-to-end model technology changes

 
As early as August 2023, Yuanrong Qixing completed road testing using an end-to-end model. In the test, the end-to-end model performed amazingly and was able to give "human intelligence" to intelligent driving cars. For example, the car will consider the needs of the car behind it. When it is parked, it will "see" the car behind it turn on its turn signal and actively let the car behind it turn right. In narrow sections of urban villages with complex road conditions, vehicles equipped with end-to-end models drive smoothly without any frustration. If there are multiple illegally parked cars ahead, the car will not park rigidly behind the car in front, but will be able to determine the driving intention of the car in front and drive around the car in front. In addition, under road conditions with obstructed vision, vehicles equipped with end-to-end models can quickly "brain-compensate" complete driving scenarios based on past training experience, so that they have the road in their "brains" and the knowledge in their "hearts".
 
 

 
Yuanrong Qixing conducts road testing on vehicles equipped with end-to-end models

 
Industry data shows that the installation rate of urban NOA functions in new energy passenger vehicles in January 2024 was 10.69%. Zhou Guang believes that at this stage, intelligent driving technology has not yet crossed the gap between the early market and the mature market, and the tacit understanding between most consumers and machines driving together needs to be improved. After the end-to-end model is put on the car, the intelligent driving system will become more human-like and easier for consumers to operate. He believes that the end-to-end model can quickly promote the intelligent driving industry towards a mature market and is an important engine for creating new productivity in the automotive industry.
 
Zhou Guang said: "The application of the end-to-end model will subvert the tool attributes of the car and turn the car into a travel robot that is good at thinking and reasoning . For the automotive industry chain, the end-to-end model will change the profit model of the car. , to increase consumers' willingness to pay for intelligent driving systems. In the future, Yuanrong Qixing will fully promote the iteration of the end-to-end model, strengthen cooperation with OEMs, and jointly create super AI drivers who understand consumers better."