Artificial intelligence E2E is gaining more ground in autonomous driving in China

From Henrik Bork* | Translated by AI 3 min Reading Time

Will 2025 be the year of E2E in autonomous driving in China? At the moment, it certainly looks like this technological route will prevail with more and more automobile manufacturers and suppliers in the Middle Kingdom.

In China, E2E is increasingly gaining ground in autonomous driving.(Image: freely licensed on Pixabay)
In China, E2E is increasingly gaining ground in autonomous driving.
(Image: freely licensed on Pixabay)

*Henrik Bork, a long-time China correspondent for the German 'Süddeutsche Zeitung' and the 'Frankfurter Rundschau', is Managing Director at Asia Waypoint, a consultancy specializing in China based in Beijing.

In "End-to-End" (E2E), data obtained from cameras or sensors—one end—are analyzed using deep learning and converted into commands for navigating, braking, or accelerating—the other end. Thanks to artificial intelligence and the increased computing power of new car chips, E2E proponents claim that autonomous driving will become "more human-like," because neural networks, just like the human brain, can not only comprehend complex situations such as construction sites, lane changes, U-turns, or potholes, but also issue smoother driving commands than previous, modularly used algorithms.

New technological revolution?

Since November 28, drivers of the AD Max model from Li Auto can activate a new, intelligent "End-to-End plus VLM" solution with a click, which has been wirelessly transmitted to their vehicle via an Over-the-Air Update (OTA). VLM stands for "Visual Language Model."

Just like several other manufacturers, Li Auto is marketing the new, heavily data and computing power-dependent driving assistance function with the label "parking lot-to-parking lot." They hope to generate interest without burdening car buyers with too many details from the world of computer science. And Li Auto is not the only manufacturer in China currently taking this path. "Intelligent driving with End-to-End has suddenly become popular," writes the Chinese business newspaper Cailianshe and asks whether E2E could trigger a new "technological revolution" in the industry even before the standard arrival of autonomous driving at Level 3.

Large models in focus

Tesla was again the pioneer in this innovation. Besides Li Auto, Xpeng and Huawei Qiankun have also launched E2E solutions on the market. Xiaomi's E2E solution is currently being tested internally. Leapmotor has also developed its own E2E system, which is expected to be available from next year. Mercedes-Benz plans to equip its new electric CLA model with E2E in China from 2025, calling it "L2++."

Software suppliers in China are also beginning to offer corresponding products. SenseAuto, the automotive business unit of SenseTime, introduced a new E2E solution for intelligent driving on November 27. Black Sesame has announced its own E2E solution for the Huashan and Wudang series of car chips. "The large models of artificial intelligence have triggered profound changes in all areas of life, including a significant improvement in production efficiency," Cailianshe quotes the CEO of SenseAuto, Wang Xiaogang. A typical example of these changes is that autonomous driving functions with E2E are currently "replacing the traditional rule-based system of autonomous driving," according to Wang. "The work of thousands of algorithm engineers writing rules can be replaced by a data-driven large model."

The more this technology is commercialized, the more data becomes available, which in turn makes E2E solutions increasingly better, argue its proponents. "The various data collected through autonomous driving is very valuable when the behavior of the vehicle changes in complex situations," says the CEO of SenseAuto.

E2E can be implemented with or without Lidar devices. This varies from manufacturer to manufacturer. While Elon Musk and Tesla rely on "Pure Vision" and initial Chinese manufacturers follow suit, as they can save on the still relatively expensive Lidars, data collected with Lidar can also be processed "end-to-end" together with data from cameras and microwave radar devices.

More intelligence

A useful feature of E2E is hard to dispute: the large models become increasingly intelligent, the more data they are allowed to "consume." Li Auto does not describe this in its marketing, but mentions to trade journalists as a result of the "rapid data processing and low latency response" of E2E models. "It is expected that the intelligent capabilities of the system will improve over time as it continuously learns and adapts to new driving conditions, making it a better and even safer driving companion." (se)

Subscribe to the newsletter now

Don't Miss out on Our Best Content

By clicking on „Subscribe to Newsletter“ I agree to the processing and use of my data according to the consent form (please expand for details) and accept the Terms of Use. For more information, please see our Privacy Policy. The consent declaration relates, among other things, to the sending of editorial newsletters by email and to data matching for marketing purposes with selected advertising partners (e.g., LinkedIn, Google, Meta)

Unfold for details of your consent