东吴汽车黄细里团队·2026-02-13 15:40Investment Highlights - The main contradiction in C-end automotive intelligence has shifted from coverage to experience optimization in 2026. Major intelligent driving manufacturers and solution providers have achieved urban NOA (Navigation on Autopilot) experiences in complex scenarios such as roundabouts and U-turns, and have improved advanced functions like parking and ETC passage. The primary optimization direction moving forward will be the handling capability of Corner Cases to enhance the driving experience for passengers and safety operators [2][7]. Road Testing Overview - This report conducted both large-sample concentrated road tests and small-sample in-depth road tests. It qualitatively and quantitatively evaluated the intelligent driving experiences of six manufacturers/solution providers: Horizon, Li Auto, Qianli Zhijia, Qingzhou Zhihang, WeRide, and Xpeng, based on dimensions such as scenario implementation, takeover frequency, and comfort. Due to subjective scales, actual road conditions, and the trust level of safety operators in intelligent driving, the report does not rank the intelligent capabilities of specific manufacturers/solution providers [3][7]. Performance Improvement - Compared to 2025, all major manufacturers have improved their intelligent driving capabilities in Q1 2026, while third-party suppliers have also demonstrated excellent implementation results. The self-research camp of manufacturers is expected to continue iterating. Li Auto and Xpeng have developed mature self-research solutions that cover full-scene NOA on public roads and some internal roads, showing improvements in static and dynamic lateral perception, scenario understanding, and Corner Case handling. Li Auto's OTA 8.2 iteration has made AD Max smoother, reducing driving jolts and stabilizing responses. Xpeng's XOS 5.8.5 version can achieve three-point U-turns and autonomously learn from parking to parking, with VLA 2.0 expected to launch after the Spring Festival, enhancing reasoning efficiency and response speed [4][7]. Third-Party Supplier Performance - Third-party intelligent driving suppliers have shown impressive results. Their solutions have been launched to cover all external public roads seamlessly, with roundabouts and U-turns executed without interruptions. Horizon's HSD demonstrated smooth performance, achieving a full hour of deep road testing with zero takeovers (with one intervention for acceleration). Qianli Zhijia's G-ASD solution on the Zeekr model significantly reduced takeover rates. Qingzhou Zhihang utilized a single J6M chip (with 128 TOPS computing power) to achieve mass production of urban NOA, which is the lowest computing power among urban NOA models. WeRide's solution won the championship at the China Intelligent Driving Competition in Taizhou, showcasing excellent lane selection logic with no safety takeovers [4][7].