Reinforcement Learning Based Manipulation Skill Transferring for Robot-assisted Minimally Invasive Surgery
Hang Su, Yingbai Hu, Zhijun Li, Alois Knoll, Giancarlo Ferrigno, Elena De Momi
Abstract
The complexity of surgical operation can be released significantly if surgical robots can learn the manipulation skills by imitation from complex tasks demonstrations such as puncture, suturing, and knotting, etc.. This paper proposes a reinforcement learning algorithm based manipulation skill transferring technique for robot-assisted Minimally Invasive Surgery by Teaching by Demonstration. It employed Gaussian mixture model and Gaussian mixture Regression based dynamic movement primitive to model the high-dimensional human-like manipulation skill after multiple demonstrations. Furthermore, this approach fascinates the learning and trial phase performed offline, which reduces the risks and cost for the practical surgical operation. Finally, it is demonstrated by transferring manipulation skills for reaching and puncture using a KUKA LWR4+ robot in a lab setup environment. The results show the effectiveness of the proposed approach for modelling and learning of human manipulation skill.
BibTeX
@inproceedings{icra2020_reinforcementlea,
title = {Reinforcement Learning Based Manipulation Skill Transferring for Robot-assisted Minimally Invasive Surgery},
author = {Hang Su and Yingbai Hu and Zhijun Li and Alois Knoll and Giancarlo Ferrigno and Elena De Momi},
booktitle = {ICRA 2020},
year = {2020}
}