Alibaba's Tongyi Lab launched the Qwen-Robot Suite in June, addressing the gap between a model that understands the physical world and one that can move something in it. The team stated that "seeing is not acting," identifying the bottleneck for embodied intelligence as the translation layer between perception and joint torques. Three new models—Qwen-RobotNav, Qwen-RobotManip, and Qwen-RobotWorld—are Alibaba's bet on closing that gap.