XPeng CEO Outlines Three Major Changes and Metrics for Autonomous Driving
XPeng CEO He Xiaopeng outlined three major structural shifts and three new core evaluation metrics for the second half of autonomous driving at the 2026 World Intelligent Connected Vehicles Conference. He also reiterated the company's plans to launch driverless Robotaxi operations next year. These insights highlight how the autonomous driving industry is transitioning from isolated software features to deeply integrated AI models, full-vehicle hardware coupling, and clearer commercialization metrics. This evolution signals a maturation phase aimed at proving real-world safety, broad consumer adoption, and sustainable business models. The three core metrics introduced for the second half of autonomous driving are safety collision mileage, user full-scenarios coverage rate, and the proportion of users willing to pay. Additionally, XPeng is utilizing its second-generation VLA model and Turing AI chips to accelerate its global Robotaxi and humanoid robot deployment.
## BACKGROUND
Autonomous driving levels range from L0 (fully manual) to L5 (fully automated under all conditions), where L4 and L5 represent highly automated or fully autonomous driving without human intervention. Robotaxis are self-driving vehicles operated for ridesharing services, while VLA (Vision-Language-Action) models expand upon vision-language frameworks to directly map visual and linguistic inputs into physical actions.