RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control

RT-2 把视觉语言模型迁移到机器人控制，实现 VLA 统一架构。