“Qwen-RobotManip uses a unified alignment framework across representation, motion, and behavioral dimensions to enable coherent, large-scale training from multiple data sources.”