Deep Reinforcement Learning-Based Control Framework for Multilateral Telesurgery
Deep Reinforcement Learning-Based Control Framework for Multilateral Telesurgery
复制标题
DOI：
10.1109/tmrb.2022.3170786
复制
发表时间：
2022-05
期刊：
影响因子：
--
通讯作者：
Sarah Chams Bacha;Weibang Bai;Ziwei Wang;Bo Xiao;E. Yeatman
中科院分区：
文献类型：
--
作者：
Sarah Chams Bacha;Weibang Bai;Ziwei Wang;Bo Xiao;E. Yeatman
传统的手术控制设计往往需要确定手术时滞的上界，这将导致跨区域手术不可行。为了克服这个问题，本文介绍了一种新的控制框架，基于深度确定性策略梯度（DDPG）强化学习（RL）算法。该框架有效地克服了时间延迟引起的相位差和数据丢失，有利于恢复外科医生的意图和交互力。卡尔曼滤波器（KF）被用来融合多个外科医生的命令，并预测最终的本地命令，分别。控制框架确保同步跟踪性能和透明度。因此，不需要时间延迟的先验知识。仿真和实验结果证明了该框架的优点。
The upper boundary of time delay is often required in traditional telesurgery control design, which would result in infeasibility of telesurgery across regions. To overcome this issue, this paper introduces a new control framework based on deep deterministic policy gradient (DDPG) reinforcement learning (RL) algorithm. The developed framework effectively overcomes the phase difference and data loss caused by time delays, which facilitates the restoration of surgeon’s intention and interactive force. Kalman filter (KF) is employed to blend multiple surgeons’ commands and predict the final local commands, respectively. The control framework ensures synchronization tracking performance and transparency. Prior knowledge of time delay is therefore not required. Simulation and experiment results have demonstrated the merits of the proposed framework.