Li Auto SVP Responds to Unitree Founder’s Criticism of VLA Model
Want to read in a language you're more familiar with?
Li Auto’s SVP of Autonomous Driving, Lang Xianpeng, issued a detailed response to Unitree founder Wang Xingxing’s criticism of the VLA (Vision-Language-Action) model, arguing that real-world large-scale data—not model architecture—is the decisive factor for embodied AI.
On December 10, Li Auto’s SVP of Autonomous Driving, Lang Xianpeng, published a long post responding to Unitree founder Wang Xingxing’s earlier doubts about the Vision-Language-Action (VLA) model.
Lang argued that model architecture is less important than its fit with an embodied-AI system, and that large-scale real-world data is the decisive factor.
He said two months of results since VLA’s official release proved that VLA is currently the best solution for autonomous driving, supported by Li Auto’s millions-vehicle data loop. Lang added that embodied intelligence ultimately depends on full-system capability, and the VLA model will serve both current vehicles and future automotive robots.
Wang previously questioned whether today’s VLA models have sufficient high-quality real-world data for reliable performance.
