Li Keqiang: Vehicle-only intelligence can't overcome safety bottlenecks; integrated vehicle-road-cloud systems are essential for advanced autonomous driving

On May 21, at the plenary session of the 13th Annual Conference on Intelligent and Connected Vehicles (ICV) held in Shanghai, Li Keqiang, Vice Chairman of the China Society of Automotive Engineers and Academician of the Chinese Academy of Engineering, delivered a speech highlighting that although Level 2 (L2) advanced driver-assistance systems (ADAS) are seeing steadily rising adoption rates and Level 3 (L3) and Level 4 (L4) autonomous driving technologies are gradually progressing toward regulatory approval and pilot demonstrations, the industry broadly faces the dilemma of “high investment but low user-perceived value,” resulting in weak revenue and gross margin performance. Li analyzed that the root cause lies in the insufficient safety and reliability of vehicle-centric intelligence in complex and critical scenarios. Current technologies struggle to achieve a fully closed-loop algorithm due to limitations such as the physical sensing constraints of onboard sensors, limited data volume within individual automakers’ closed-loop systems, inconsistent data formats, and spatiotemporal information gaps caused by occlusions. Therefore, he argued, autonomous driving development must shift toward a new paradigm: vehicle intelligence remains the foundation, but integration of vehicles, roads, and cloud systems—“vehicle-road-cloud integration”—is its inevitable evolutionary path. Vehicle-road-cloud integration creates a “bird’s-eye-view” digital infrastructure capable of effectively addressing critical scenarios that are difficult for individual vehicles to handle alone, such as sudden pedestrian crossings (“ghost probing”) and ramp merging. Moreover, this integrated system aggregates comprehensive spatiotemporal data, providing automotive AI large models with richer training datasets and beyond-line-of-sight reference inputs, thereby helping resolve longstanding challenges in industrial control related to real-time performance and interpretability. To date, 21 leading domestic and international automakers have joined collaborative demonstration projects for vehicle-road-cloud integration, with seven already exploring mass production. Applications span 17 representative scenarios, including green-wave speed guidance, ramp-merge assistance, and collective safety warnings, and have been validated in cities such as Beijing, Chongqing, and Shanghai. This approach emphasizes standardized protocols and cost-effectiveness and is fostering a new business model involving automakers, platform providers, and infrastructure operators. Li stressed that future high-reliability, high-level autonomous driving must rely on the physical-data foundation enabled by vehicle-road-cloud integration to achieve full synergy among humans, vehicles, roads, and the cloud—rather than depending solely on competition centered on individual vehicles.