65% L2 ADAS Adoption Still Unprofitable as Safety Bottlenecks Hinder Scale-Up

At a recent high-level forum on the development of intelligent electric vehicles, Li Keqiang, Professor at Tsinghua University and Chief Scientist of the National Innovation Center for Intelligent and Connected Vehicles, stated that although the intelligent driving industry has made significant progress, it still faces core challenges such as insufficient safety and reliability and the absence of a viable commercial closed loop. Data shows that by 2025, the adoption rate of Level 2 (L2) advanced driver-assistance systems (ADAS) in new vehicles has already approached 65%, yet the industry as a whole remains unprofitable. Li Keqiang analyzed that key technical bottlenecks hindering large-scale deployment include: physical blind spots in vehicle-based perception systems, which struggle to handle complex scenarios like "jaywalking pedestrians" ("ghost probing") and merging onto ramps; incomplete and inaccurate training data for end-to-end large models, leading to inadequate reliability; and fragmented, siloed ("chimney-style") architectures adopted by different companies, resulting in high development costs, low efficiency, and redundant resource investment. He proposed that an integrated "vehicle-road-cloud" architecture is the critical pathway to overcoming these challenges. This approach enables collaborative perception among vehicles, road infrastructure, and cloud platforms, establishing a comprehensive digital framework to enhance environmental awareness and risk prediction capabilities. Currently, this system is undergoing industrial-scale demonstrations with 15 leading domestic and international automakers, seven of which are advancing toward mass production. In cities such as Beijing and Chongqing, vehicles have already achieved green-wave passage through more than five consecutive traffic-light-controlled intersections and have successfully validated performance in 17 typical scenarios, including connected forward collision avoidance, ramp merging, and "ghost probing" warnings. Li Keqiang emphasized that the industry must return to its foundational priority—safety—and abandon short-termism. He advocated adopting a systems engineering mindset to drive technological breakthroughs. The "vehicle-road-cloud" integrated architecture can aggregate massive, comprehensive, multi-source data to support efficient training of AI large models, while layered decoupling and low-code development platforms can significantly improve R&D efficiency and product safety.