XPeng Unveils TuringViT Efficient Vision Encoder to Power Autonomous Driving, Smart Cockpit, and Humanoid Robotics

On July 21, XPeng Group unveiled TuringViT, a highly efficient visual encoder. Designed for the era of Vision-Language Models (VLMs) and Vision-Language-Action models (VLAs), this encoder features a systematic redesign of visual encoder architecture, data paradigms, and training pipelines. It will be comprehensively deployed across three core business scenarios: intelligent driving, smart cockpits, and IRON humanoid robots.