Learning Topological Motion Primitives for Knot Planning

Abstract:

Motion planning for knot tying is a challenging problem. We propose to learn a library of topological motion primitives that can be composed to create various knots. Knotting tasks are decomposed into sequences of abstract topological actions based on knot theory. For each topological action, we train a motion primitive that generates robot motion trajectories conditioned on the observed rope configurations. We train the motion primitives by imitating human demonstrations and reinforcement learning in simulation. To generalize human demonstrations of simple knots into more complex knots, we observe similarities in the motion strategies of different topological actions, design the neural network structure and inputs to exploit such similarities. We demonstrate that our learned motion primitives can be used to efficiently generate motion plans for tying the overhand knot, and the motion plan can be executed on a real robot using visual tracking and Model Predictive Control. We also demonstrate that our learned motion primitives can plan for a more complex pentagram-like knot, when the human demonstrations are only on simpler tasks.

[RSS 4-page abstract] [IROS full paper] [code]

BibTex:

@inproceedings{yan2020TMP,

title={Learning Topological Motion Primitives for Knot Planning},

authors={Mengyuan Yan and Gen Li and Yilin Zhu and Jeannette Bohg},

booktitle={IEEE International Conference on Intelligent Robots and Systems (IROS)},

year={2020}

}