Flow as Flow: Modeling Robot Velocity Fields as Probability Velocity Fields for Flow-Based Object Manipulation
2026-06-22 • Robotics
Robotics
AI summaryⓘ
The authors focus on improving how robots learn to move objects by representing robot motions as continuous flows of velocity, rather than just a few key movement points. They introduce a new method called Flow as Flow that models these velocity fields using a probabilistic approach, leading to faster and better robot motion generation. Their experiments show that this method works much faster and has higher success rates than previous techniques on many tasks. They also tested it extensively in real-world scenarios and found it more reliable than other methods.
robotic foundation modelsobject manipulationrobot velocity fieldsflow matchingprobability flowsdense velocity fieldsrobot motion generationbenchmark evaluationcontinuous-time motionreal-world robotic experiments
Authors
Koki Seno, Daichi Yashima, Yusuke Takagi, Kento Tokura, Komei Sugiura
Abstract
Cross-embodiment data have become central to training robotic foundation models. To leverage such heterogeneous data, we focus on flow-based object manipulation, where robot flows (robot velocity fields) serve as embodiment-agnostic motion representations. Previous studies do not formulate robot flows as dense velocity fields, but as displacements of sparse keypoints, while such velocity fields better match the continuous-time nature of motions. We propose Flow as Flow, a framework that models robot flows as probability flows based on a flow matching formulation. By naturally modeling such velocity fields within this formulation, our method achieves efficient and high-quality robot flow generation. Across standard benchmarks, our method outperforms representative baseline methods on standard metrics, while achieving approximately 33$\times$ faster generation. Furthermore, through real-world experiments evaluating 9 methods with 260 trials per method across 13 manipulation tasks, we show that our method achieves a higher average success rate than the baseline methods. Our project page is available at https://flow-as-flow-u0n5y.kinsta.page.