Qwen-Robot

Qwen's entry into robotics: three simultaneous technical reports, with weights explicitly not released. RobotNav is a navigation foundation model with a parameterized inference-time observation interface, trained on 15.6M samples with favorable 2B→8B scaling and zero-shot real-robot transfer. RobotManip is a vision-language-action model built on Qwen-VL over a ~38,100-hour pretraining corpus including human-to-robot data synthesis across 15 robot platforms. RobotWorld is a language-conditioned video world model unifying manipulation, driving, and navigation.