Chinese electric vehicle maker Li Auto is revving up its efforts to accelerate its autonomous driving technology development. This move could close the gap between Li Auto and industry leader Tesla.
Li Auto has now shifted its focus to an end-to-end (E2E) approach enhanced by a vision language model (VLM). This mirrors Tesla’s ‘One Model’ design more closely than its domestic rivals in the US.
Li Auto chasing Tesla
Lang Xianpeng, vice president of smart driving R&D at Li Auto, explained the company’s strategy in an interview with KR-Asia. He says “E2E marks the first use of AI in smart driving.”
Lang explains that the E2E system eliminates separate modules for perception, prediction, planning, and control. This essentially merges the different modules into a single neural network.
The company combines E2D with a VLM and what they call a ‘world model.’
Concept clarified: The VLM acts as a plugin similar to ChatGPT. It can provide cognitive and logical reasoning capabilities. In complex scenarios, the system can query the VLM in real-time for driving assistance.
A Vision Language Model integrates visual data processing with language understanding to enhance decision-making processes in autonomous systems.
Bringing AI to driverless cars
Jia Peng, head of smart driving technology R&D at Li Auto, says the VLM is the core of Li Auto’s smart driving system.
He explains: “VLM itself is also following the development of large language models (LLMs), and no one can yet answer how large the parameter count will eventually be.”
That said, Li Auto is investing large amounts of money in computational power to support its AI-driven approach. According to Lan, the company’s cloud computing power is 4.5 EFLOPS, and it’s rapidly closing the gap with other industry giants, like Huawei.
EFLOPS explained
EFLOPS is a unit of measurement; it stands for Exa Floating Point Operations Per Second.
1 EFLOPS is equivalent to 1 quintillion floating-point operations per second, so 4.5 EFLOPS is 4,500,000,000,000,000,000 FLOPS.
To give you an idea of how massive this is, the fastest personal computers operate in the range of gigaflops (GFLOPS). And 4.5 EFLOPS is a trillion times more powerful than that.
Tesla in the lead, but for how long?
Li Auto’s CEO Li Xiang says the company is “fully committed” to its strategy. He jokes that whenever Lang asks him if they have computational power, he’ll say to “source more from Xie Yan, the company’s CTO.”
Xiang also admits that Tesla is still the industry leader, but is proud to say that Li Auto is “catching up quickly.” He says: “”From a product experience perspective, we’re at the level Tesla was when it released its E2E version last year—about a six-month gap.”
With this vast amount of computational power behind Li Auto, how will Tesla still be in the lead?
ALSO READ: Tesla Robo-taxi reveal delayed due to ‘important design change’
Tesla vs Li Auto: Closing the gap
Meanwhile, Li Auto is learning from Tesla. Jia says Tesla’s biggest issue at the moment is validation. “FSD V12.4 performed poorly, leading to V12.5’s fivefold parameter increase. I suspect validation was lacking, not knowing how it performs for users,” Jia explains.
Li Auto will continue to focus on developing a “robust world model for validation,” as well as a more advanced VLM which may surpass Tesla.
Jia notes: “These insights led us to develop VLM, aiming to break this ceiling. VLM’s high potential offers a path to surpass Tesla.”
As the race for autonomous driving heats up, Li Auto’s aggressive investment in AI and computational power positions it as a serious contender in the chase to catch Tesla.
NOW READ: Advancements in autonomous vehicles for logistics and delivery
Image credit: lixiang
Share this article
About the author
Cheryl has contributed to various international publications, with a fervor for data and technology. She explores the intersection of emerging tech trends with logistics, focusing on how digital innovations are reshaping industries on a global scale. When she's not dissecting the latest developments in AI-driven innovation and digital solutions, Cheryl can be found gaming, kickboxing, or navigating the novel niches of consumer gadgetry.
