Chu Wenbo: Further integrate AI big models and optimize core technologies for autonomous driving
2024-08-19
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
China Youth Daily Client News (China Youth Daily·China Youth Network Intern Reporter Jia Jiye Reporter Zhang Miao) "Further integration of AI big models has become the key to solving the core barriers of 'end-to-end' autonomous driving." Chu Wenbo, general manager of the Intelligent Connected Vehicle Innovation Center of China Western Science City and chief technical expert of the National Intelligent Connected Vehicle Innovation Center, said at the 2024 World Youth Development Forum Digital Development Theme Forum held on August 15.
"End-to-end" is an autonomous driving technology. Its core is to input road, vehicle, audio-visual and other perception information into the autonomous driving system, and then directly output decision commands. The intermediate process is completed automatically by artificial intelligence, realizing the integration of perception, decision-making, planning and other modules in the traditional autonomous driving system.
In Chu Wenbo's view, the emergence of large models promotes the evolution of autonomous driving from modularization to "end-to-end". However, he also said that "end-to-end" autonomous driving still faces great problems. "End-to-end" autonomous driving is highly dependent on training data, and the currently available data has problems such as low realism of simulation scenarios, poor generalization of training scenarios, and single labeled data type. In addition, the "end-to-end" reasoning process is a hidden "black box", and it is difficult to supervise the variables in the decision-making process, which can easily lead to a series of safety risks.
"We need to further integrate AI big models to optimize the core technology of 'end-to-end' autonomous driving." Chu Wenbo said that first, we need to use the world model (one of the key paths to general artificial intelligence) to enhance driving perception. In addition to the surrounding lane lines, this technology can even perceive more realistic driving conditions including weather, light and shadow, and shape; secondly, we need to use the information of the multimodal big model. In addition to the information provided by the vehicle sensors, we also need to add information from other clouds. At the same time, we can also realize driving interaction through the visual language big model, so that the vehicle can form a series of text information according to the driving situation to directly guide the vehicle's movement decision.
In addition, Chu Wenbo also mentioned the need to "solidify safety", that is, strive to achieve the explainability of "end-to-end" autonomous driving to reduce driving risks. On the one hand, the controllability of the model should be improved, and on the other hand, more intermediate links can be output to facilitate safety monitoring.
On August 15, the 2024 World Youth Development Forum Digital Development Theme Forum was held in Chongqing. Chu Wenbo, general manager of the Intelligent Connected Vehicle Innovation Center of Western China Science City and chief technical expert of the National Intelligent Connected Vehicle Innovation Center, was delivering a keynote speech. Photo provided by the organizer
Source: China Youth Daily Client
