Huawei Unveils Full-Stack ADS 5 Solution, Systematically Accelerating L3 Autonomous Driving at Scale

Huawei held a technology conference on the eve of the 2026 Beijing Auto Show, officially unveiling its Qiankun ADS 5.0 intelligent driving system and simultaneously launching its self-developed Qiankun OS operating system. Centered around two core pillars—safety and architecture—the solution establishes a complete closed-loop system spanning from cloud-based training to vehicle-side execution, aiming to provide full-stack support for the large-scale commercialization of Level 3 (L3) autonomous driving. At the algorithmic level, ADS 5.0 introduces a cloud-based world model grounded in multi-agent game theory, integrated with the WEWA 2.0 architecture to enable online reinforcement learning that “generates, learns, and validates” concurrently—boosting training efficiency by 10x. On the vehicle side, it applies for the first time the Safety Risk Field theory, which quantifies kinetic fields, potential fields, and behavioral fields to generate dynamic risk heatmaps, reportedly reducing collision risk by 50%. The Qiankun OS features a deterministic low-latency engine, an end-to-end zero-trust security model, and a comprehensive safety redundancy architecture. Leveraging technologies like the Lingqu Bus, it reduces in-vehicle signal latency by 30%, ensuring timely responses in emergency scenarios. In terms of safety architecture, ADS 5.0 upgrades from the original “five-dimensional safety” framework to a “six-dimensional safety” system, adding a new dimension: “full-time-domain safety.” This covers pre-event warnings (e.g., hazard alerts within five kilometers, proactive speed reduction at intersections when speeding) and post-event protections (e.g., third-row seat联动 protection during rear-end collisions, Blowout Stability Control Assist 3.0, and Driver Incapacitation Assist 2.0). As of April 19, 2026, Qiankun Intelligent Driving has accumulated over 10 billion kilometers of assisted driving mileage, and Huawei announced it will publicly display this data in real time on its official website. Technologically, Huawei explicitly endorses a multi-sensor fusion approach and opposes pure vision-only solutions, opting for a World Model (WA) path rather than a Vision-Language-Action (VLA) architecture. Commercially, Huawei positions itself as an “electronic screw,” focusing solely on empowering automakers without engaging in whole-vehicle manufacturing. To date, Qiankun has partnered with more than 25 brands and 50 vehicle models, with over 1.7 million units deployed. In ecosystem expansion, ADS 5.0 also upgrades its Point-to-Point Parking 3.0 feature, planning to connect by Q2 2026 to 300,000 parking lots, 1.8 million charging piles, and 3,000 car wash stations—extending intelligent driving services into everyday life scenarios. Huawei expects to invest over RMB 18 billion in Qiankun system R&D in 2026, a figure it claims exceeds the combined R&D spending of all other major domestic solution providers.