At the Second Autonomous Driving Industry Development Forum held in Beijing on April 25, Liu Bin, Chief Designer Expert at the R&D Academy of China FAW Group Co., Ltd., stated that autonomous driving technology solutions have not yet truly converged. Each participant in the industry has its own investment priorities and approaches. Finding common ground to avoid redundant investments across the industry is crucial.
From a model composition perspective, there has been a rapid evolution from traditional Software 1.0, which is based on code and rule orchestration, to data-driven Software 2.0. With enhanced AI capabilities, the evolution cycle of models in the 3.0 era continues to shorten, which is a broad characteristic of the new age.
What is the core from a data perspective? Liu Bin pointed out that it is the so-called closed loop—how to build a closed loop from the technology itself to the entire industry chain ecosystem is a key aspect of data as a foundational element.
In his view, from a model perspective, the most important current focus is the evolution or generalization capability of AI models. Today's AI models are not only tasked with fulfilling the mission of autonomous driving but also need to integrate extensively with the cockpit and other vehicle domains. Only in this way can we achieve an understanding of the true laws of the physical world, realize a high degree of unity between semantics and physical comprehension, and attain higher performance ceilings and stronger generalization capabilities.
Furthermore, Liu Bin emphasized that computational power must ultimately serve as the foundation. He noted that a key requirement for FAW regarding computational power is controllability. This year, at the auto show, FAW launched the Hongqi No. 1 integrated autonomous cockpit and driving chip, which not only supports autonomous driving and cockpit functions but also incorporates vehicle control computing into the chip, enabling full-domain development and controllability.
Comments