Learning Arbitrary-Goal Fabric Folding with One Hour of Real Robot Experience

Manipulating deformable objects, such as fabric, is a long standing problem\nin robotics, with state estimation and control posing a significant challenge\nfor traditional methods. In this paper, we show that it is possible to learn\nfabric folding skills in only an hour of self-supervised real robot experience,\nwithout human supervision or simulation. Our approach relies on fully\nconvolutional networks and the manipulation of visual inputs to exploit learned\nfeatures, allowing us to create an expressive goal-conditioned pick and place\npolicy that can be trained efficiently with real world robot data only. Folding\nskills are learned with only a sparse reward function and thus do not require\nreward function engineering, merely an image of the goal configuration. We\ndemonstrate our method on a set of towel-folding tasks, and show that our\napproach is able to discover sequential folding strategies, purely from\ntrial-and-error. We achieve state-of-the-art results without the need for\ndemonstrations or simulation, used in prior approaches. Videos available at:\nhttps://sites.google.com/view/learningtofold\n