On August 31st, Red Star Capital Bureau reported that the 27th Chengdu International Auto Show kicked off on August 30th. Ideal Automobile (02015. HK/LI. US) announced the latest progress and future plans of end-to-end model, VLM visual language model, and world model autonomous driving technology architecture, and announced the recruitment of a new generation of Ideal Intelligent Driving based on end-to-end and VLM visual language models for a 10000 person experience group. In addition, Ideal OTA6.2 has officially been fully launched.
Lang Xianpeng, Vice President of Intelligent Driving Research and Development at Ideal Automobile, and Zhan Kun, Senior Algorithm Expert in Intelligent Driving, were interviewed by media groups including Red Star Capital Bureau after the press conference.
According to Zhan Kun, end-to-end refers to a paradigm of research and development, where a task is completed from the initial input to the final output without any other processes, using a single model to complete the entire process from input to output. Applied to the field of autonomous driving, it means that with just one model, the perception information collected by sensors such as cameras can be converted into vehicle operation instructions.
At the beginning of 2023, Tesla (TSLA. US) mentioned end-to-end. At present, many car companies are rolling end-to-end, but their ideas and progress are different. In Zhan Kun's view, compared to modular end-to-end, OneModel end-to-end is more fundamental.
Lang Xianpeng said, "Our core idea for intelligent driving is end-to-end+VLM. We believe that this approach is a more promising intelligent driving solution that is closer to human driving
In Lang Xianpeng's view, the end-to-end+VLM technology architecture is essentially an artificial intelligence solution. From now on, we are truly using artificial intelligence to do autonomous driving. He believes that under this premise, the core competition in autonomous driving research and development is whether there is more and better data and matching computing power to train models, and training data and training mileage cannot be bought with money.
He revealed that the current training computing power of Ideal Automobile has reached 5.39 EFLOPS, and it is expected to exceed 8 EFLOPS by the end of 2024. Ideal Auto invests over 1 billion yuan annually in training computing power, and is expected to consume 2 billion yuan this year. We believe that the training computing power required to ultimately achieve autonomous driving should reach the level of 100 EFLOPS, which translates to an annual investment of over 1 billion US dollars
Regarding the world model, Lang Xianpeng pointed out that in supervised (L3 level and below) autonomous driving, end-to-end models and VLM visual language models still play a greater role, "because under the demand for supervised autonomous driving, end-to-end models are already sufficient, and VLM only serves as a reminder and assistance. But after reaching unsupervised L4 level autonomous driving, this system has to independently handle all unknown scenarios and unexpected situations, and the number of model parameters increases dramatically. At this time, a world model on the vehicle side is needed.
Takeover rate is one of the core indicators that reflect the system's capabilities. It is reported that Ideal Auto only takes over once after reaching over 21 kilometers. In the future, it can be upgraded to take over every 100 kilometers. However, if there is no need to take over for a long time, people's mental attention will not be focused. Ideal cars will provide a brand new interactive experience, allowing drivers to take over when necessary, and by evaluating high-frequency takeover scenarios and regions, they will be pushed to users in advance.
Compared to its competitors, Ideal Auto currently does not charge for advanced intelligent driving. Lang Xianpeng emphasized that both standard and free are strategies developed from the first day of entering intelligent driving, and supervised autonomous driving is free for all AD Max owners. Delivery volume is a very important measurement indicator. For us, it is not just about delivery volume, but also about providing more vehicle training mileage for autonomous driving. Delivery volume is relatively good and the enterprise operates steadily, and there are sufficient resources to invest in intelligent driving research and development
Last year, Lang Xianpeng stated that the gap between Ideal Auto's intelligent driving and Tesla FSD was about six months. At this year's Chengdu Auto Show, he said that "the gap between the two sides may be even smaller this year".
He explained that firstly, in terms of technical architecture, Ideal Auto is not much different from Tesla, and even more advanced, "because we have VLM and System 2, Tesla only has System 1, end-to-end".
Secondly, in terms of training computing power and training data in China, "at least from now on, we are ahead of Tesla because Tesla is constrained in data compliance and other aspects, and the deployment of training computing power in China still needs to be established. In this regard, the gap between us and Tesla in China may not be so big, and we also hope that Tesla can join in, learn from each other, and focus on improving ourselves