Image Source: China Visual

BEIJING, June 19 (TMTPost)—— Li Auto plans to roll out its assisted-driving system on urban roads in Beijing and Shanghai at the end of June, according to the announcement released by the company last Saturday.

The company will begin with a beta test of its NOA system — Navigation on Advanced Driver Assistance Systems — with a small group of users in the two cities.

In the second half of the year, it will also launch smart driving for commuting, which allows urban drivers nationwide to use the assisted-driving system on a specific route before the NOA feature is released in their cities.

As the most advanced stage of smart cars, autonomous driving will overturn the existing car business model once it is realized. BYD Chairman Wang Chuanfu noted at the annual report communication meeting that driverless car was a gimmick at the end of March. His comments sparked controversy but also reflected that the industry starts to be rational in viewing autonomous driving technology.

The industry consensus is that autonomous driving still needs human driver assistance and supervision. In order to avoid consumer confusion about the system's capabilities, companies commonly rename “autonomous driving systems” as “intelligent driving systems” and also divide use cases into the three main categories of parking, highway driving, and city driving.

Among the above categories, realizing intelligent driving in cities is the most difficult and valuable task. According to Lang Xianpeng, the vice president of intelligent driving at Li Auto, the industry's biggest pain point at the moment is that consumers do not have an intuitive perception of the value of intelligent driving. It is mainly because car companies have just achieved high-speed assisted intelligent driving, but consumers spend most of their time driving vehicles for trips within the city.

Data from the Ministry of Transport show that China's total road length was about 5.35 million kilometers by the end of 2022, including 177,000 kilometers of highways. Lang said that 82% of the miles covered by Li Auto users were in the city.

In 2020, Tesla took the lead in launching a beta version of its assisted intelligent driving system FSD (Full-Self Driving) for urban scenarios.

The technology route that Tesla takes is not easy. Ai Rui, vice president of self-driving technology company Haomo.AI said that autonomous driving previously used a variety of deep learning network architectures, but after the emergence of Transformer in 2017, these different architectures converge to the same form as Transformer, which "can be understood as convergence at the algorithm level."

Transformer is a network architecture proposed by Google Scholar in a paper published in 2017. Ai believes that the introduction of the Transformer architecture means that the development of the autonomous driving industry has shifted from relying on genius engineers to relying on data and computing power.

The competition between companies for data and computing power implies the start of the "arms race" in this industry. Tesla released its own training chip D1 in September 2021, which it planned to use to form a supercomputer group Dojo ExaPOD, with a total computing power of 1.1EFLOPS. Tesla planned at the time that the supercomputing cluster would be deployed in the first quarter of 2023, with the long-term goal of deploying seven supercomputing clusters in California.

本文内容来源：钛媒体英文站