Li Xiang reveals new breakthroughs in autonomous driving technology: innovation in simulating human driving methods

At the "2024 China Automotive Chongqing Forum", Li Xiang, founder of Ideal Auto, revealed to the participants the major breakthroughs in autonomous driving technology achieved by the Ideal Auto team in the past six months. The core concept of this technology is to simulate human driving methods to achieve more intelligent and efficient autonomous driving.

Li Xiang first emphasized the difference between "autonomous driving" and "intelligent driving" or "assisted driving". He pointed out that the core of autonomous driving is to completely replace human control of the vehicle, rather than just providing assistance or intelligent suggestions. In order to achieve this goal, the Ideal Auto team has been focusing on studying the essence of human driving since September last year and trying to incorporate this essence into autonomous driving technology.

Li Xiang mentioned that humans do not need to constantly learn how to operate a vehicle when driving, but rely on intuition and experience to deal with various road conditions. How are these intuitions and experiences formed? He quoted the theory in the book "Thinking, Fast and Slow", pointing out that the human brain is divided into System 1 and System 2 when processing information. System 1 is responsible for fast, intuitive reactions, similar to most situations when humans drive; while System 2 is responsible for dealing with complex problems that require logical deduction, such as driving decisions in special situations.

Based on this theory, the Ideal Auto team began to explore how to apply this human driving style to autonomous driving technology. They proposed the concepts of System 1 and System 2 for autonomous driving. System 1 corresponds to end-to-end autonomous driving technology, that is, an efficient mode of direct input and output, avoiding the cumbersome process of traditional perception, planning, execution and other modules. However, end-to-end technology also faces challenges, such as the need for professional talents, high-quality data and strong computing power support.

To meet these challenges, the Ideal Auto team further explored the application of System 2. System 2 is inspired by how humans improve their abilities when facing complex problems rather than just solving problems through learning. Take Li Xiang's wife's driving experience as an example. After participating in BMW driving training to improve her ability to see the road and brake, her driving skills have been significantly improved.

In the field of autonomous driving, System 2 aims to solve complex and generalized problems that end-to-end technology cannot handle. To this end, Ideal Auto introduced a visual language model (VRM), which can read navigation maps and make predictions like humans. VRM not only provides a bottom-line guarantee for end-to-end technology, but also solves driving problems under various complex road conditions. It is worth mentioning that VRM also enables the autonomous driving system to say goodbye to its dependence on high-definition maps, which is a major technological breakthrough.

Li Xiang revealed that their autonomous driving system uses two chips, one for running end-to-end technology and the other for running a compressed VRM model. The verification results of this system are exciting, and they expect to achieve truly supervised L3 autonomous driving as early as the end of this year or at the latest in the first half of next year.

However, as AI technology is increasingly used, how to verify AI capabilities has become a new challenge. The Ideal Auto team proposed a new verification method, which is to use a small visual model to simulate the human test process to verify the capabilities of the autonomous driving system. This verification method not only conforms to the working principles of humans, but also provides new ideas for the development of autonomous driving technology.

Looking ahead, Li Xiang said that Ideal Auto will launch the map-free NOA function in the third quarter of this year, and plans to gradually launch an end-to-end + VRM supervised autonomous driving system in the future. With the continuous evolution of technology and the increase in computing power, he firmly believes that it is completely feasible to achieve supervised L3 autonomous driving on the existing computing platform, and expects to achieve unsupervised L3 autonomous driving within three years.

Li Xiang's sharing not only demonstrated their major breakthroughs in autonomous driving technology, but also provided new ideas and directions for the development of the entire industry. With the continuous advancement of technology and the expansion of application scenarios, we have reason to believe that autonomous driving technology will bring people a more convenient and safer travel experience in the future.