Qwen-VLA: Unifying Vision-Language-Action Modeling across Tasks, Environments, and Robot Embodiments

Qwen-VLA is a unified vision-language-action foundation model for robotics that turns visual and language understanding into continuous robot actions.

It is jointly pretrained on robotics, navigation, simulation, and auxiliary vision-language data, with embodiment-aware prompts to support different robot platforms.

The model performs strongly on manipulation, navigation, and trajectory benchmarks, including LIBERO, Simpler-WidowX, RoboTwin, R2R, RxR, ALOHA, and DOMINO.

Its out-of-distribution robustness suggests a more general robot AI system that could transfer better across tasks, environments, and embodiments.