Autonomous Tissue Manipulation via Surgical Robot Using Deep Reinforcement Learning and Evolutionary Algorithm

Abstract

Robotic surgery makes use of autonomous robots that can perform some surgical tasks on their own. Surgical robots performed well in conjunction with machine learning, particularly reinforcement learning (RL), allowing them to be used in complex environments, such as cutting a pre-determined pattern on soft tissue with surgical scissors and gripper. There is no doubt that soft tissue is deformable, so using a tensioning policy can determine appropriate tension direction from the pinch point at any time to have an accurate cut in the pre-determined trajectory. In this study, we used the deep reinforcement learning (DRL) approach to find an optimal tensioning policy for cutting soft tissues. In addition, we used an evolutionary algorithm with the operators appropriate to the problem and the learned tensioning policy to find the sequence of tensioning actions. The objective of this study is to determine the optimal tensioning policy and the best tensioning action sequence. The experimental results show that using the learned policy results in smaller damage and error and lead to the highest scores compared to the previous studies.