Huawei Executive Says Pure Vision Approach Still Far from L4 Autonomy

At the 13th Annual Conference on Intelligent Connected Vehicles held in Shanghai on May 21, Li Wenguang, President of Huawei's ADS (Autonomous Driving Solution) product line, delivered a speech stating that intelligent connected vehicles are currently transitioning from "human-machine co-driving" to higher-level autonomous driving. Based on over 11.1 billion kilometers of accumulated intelligent driving mileage data, the latest ADS 4.1 version has reduced the severe accident rate to once per 7.2 million kilometers—4.37 times safer than the average Chinese driver. Moreover, its fully unmanned valet parking (VPD) solution deployed in campuses for over a year has demonstrated safety performance 30 times better than human drivers. Li emphasized that these achievements fall within the scope of human-machine co-driving and do not indicate that Level 4 (L4) autonomous driving is already mature. He explained that L4 systems must achieve million-kilometer-level stability and deterministic scheduling capabilities in their operating systems under unsupervised conditions. He clearly stated that pure vision-based solutions still exhibit significant gaps compared to L4 requirements in extreme combined scenarios such as nighttime driving, dark-colored objects, backlighting, and heavy rain with sensor contamination. Additionally, Li expressed opposition to adopting an integrated "cockpit-and-driving" architecture in premium vehicles, noting fundamental differences between intelligent cockpits and autonomous driving systems in terms of computing power requirements, operating systems, and safety standards. While integrating both functions onto a single chip may slightly reduce packaging costs, it lowers manufacturing yield and offers limited overall benefits. Furthermore, current software technology lacks the capability to run two distinct operating systems simultaneously on one chip. According to Li’s forecast, Level 3 (L3) autonomous driving will begin pilot deployments in 2026, high-level intelligent driving will achieve large-scale commercialization in 2027, and Robotaxis, unmanned long-haul logistics, and full-speed urban L4 services will see commercial rollouts in select cities by 2028.