NovaFlow was jointly proposed by a research team from Brown University and the Institute for Robotics and Artificial Intelligence in October 2025, and the relevant research results were published in the paper "NovaFlow: Zero-Shot Manipulation via Actionable Flow from Generated Videos".

NovaFlow is an object-centric, embodiment-free autonomous manipulation framework that translates task descriptions into executable plans for a target robot without any demonstration. Given a task description, the framework synthesizes video using a video generation model and refines it into a stream of 3D manipulable objects using readily available perception modules. By decoupling task understanding from low-level control, NovaFlow enables natural cross-embodiment transfer. Researchers validated its implementation on rigid, articulated, and deformable object manipulation tasks using the desktop Franka robotic arm and the Spot quadrupedal mobile robot, achieving efficient zero-shot execution without demonstrations or embodiment-specific training.