RoboDream: Compositional World Models for Scalable Robot Data Synthesis

ArXi:2606.02577v1 Announce Type: cross Scaling robot learning requires large-scale, diverse nstrations, yet real-world data collection via teleoperation remains prohibitively expensive and time-consuming. While video diffusion models offer a promising avenue for data scaling, existing generative approaches are often limited to superficial visual augmentation, or suffer from embodiment hallucinations that yield physically infeasible motions.