FabricFlowNet: Bimanual Cloth Manipulation with a Flow-based Policy
Thomas Weng, Sujay Bajracharya, Yufei Wang, Khush Agrawal, David Held
Robotics Institute, Carnegie Mellon University, Pittsburgh, PA
[arXiv] [OpenReview] [Poster] [Code]
This work as has been accepted at CoRL 2021.
Abstract
We address the problem of goal-directed cloth manipulation, a challenging task due to the deformability of cloth. Our insight is that optical flow, a technique normally used for motion estimation in video, can also provide an effective representation for corresponding cloth poses across observation and goal images. We introduce FabricFlowNet (FFN), a cloth manipulation policy that leverages flow as both an input and as an action representation to improve performance. FabricFlowNet also elegantly switches between dual-arm and single-arm actions based on the desired goal. We show that FabricFlowNet significantly outperforms state-of-the-art model-free and model-based cloth manipulation policies. We also present real-world experiments on a bimanual system, demonstrating effective sim-to-real transfer. Finally, we show that our method generalizes when trained on a single square cloth to other cloth shapes, such as T-shirts and rectangular cloths.
Videos of FabricFlowNet Executing Square Cloth Goals
Our method is trained in simulation and transferred to the real world.
Goal (All Corners In)
Achieved
Goal (Opposite corners in)
Achieved
Goal (Test 26)
Achieved
Goal (Two side horizontal)
Achieved
Goal (Triangle)
Achieved
Goal (Test 6)
Achieved
Goal (Test 7)
Achieved
Goal (Test 8)
Achieved
Goal (Test 9)
Achieved
Goal (Test 10)
Achieved
Goal (Test 27)
Achieved
Goal (Test 11)
Achieved
Goal (Test 24)
Achieved
Goal (Test 25)
Achieved
Goal (One corner in)
Achieved
Goal (Double Rectangle)
Achieved
Goal (Double Triangle)
Achieved
Videos of FabricFlowNet Generalizing to Rectangular Cloth and T-Shirt Goals
Our method is trained only on a square cloth in simulation and generalizes to other cloth shapes.
Goal (T-shirt horizontal)
Achieved
Goal (T-shirt vertical)
Achieved
Goal (T-shirt full)
Achieved
Goal (Rectangle two side horizontal)
Achieved
Goal (Rectangle one corner)
Achieved
Goal (Rectangle two side vertical)
Achieved
Goal (Rectangle horizontal)
Achieved
Goal (Rectangle vertical)
Achieved
Acknowledgements
This work was supported by the National Science Foundation (NSF) Smart and Autonomous Systems Program (IIS-1849154), a NSF CAREER Award (IIS-2046491), LG Electronics, and a NSF Graduate Research Fellowship (DGE-1745016).