Rearranging Deformable Linear Objects for Implicit Goals with Self-Supervised Planning and Control

The robotic manipulation of deformable linear objects is a frontier problem with many potential applications in diverse industries. However, most existing research in this area focuses on shape control for a provided explicit goal and does not consider physical constraints, which limits its applicability in many real-world scenarios. In this study, a self-supervised planning and control approach are proposed to address the challenge of rearranging deformable linear objects for implicit goals. Specifically, the context of making both ends of the object reachable (inside the robotic access range) and graspable (outside potential collision regions) by dual-arm robots is considered. Firstly, the object is described with sequential keypoints and the correspondence-based action is parameterized. Secondly, a generator capable of producing multiple explicit targets is developed, which adhere to implicit conditions. Thirdly, value models are learnt to assign the most promising explicit target as guidance and determine the goal-conditioned action. All models within the policy are trained in a self-supervised manner based on data collected from simulations. Importantly, the learned policy can be directly applied to real-world settings since we do not rely on accurate dynamic models. The performance of the new method is validated with simulations and real-world experiments.