Path Planning for Automation of Surgery Robot based on Probabilistic Roadmap and Reinforcement Learning

Abstract

Laparoscopic robotic surgery is a new surgical method performed by inserting several surgical tools and a laparoscope through an umbilical incision [12]. Compared with conventional laparoscopic surgery minimizes patient pain with minimally invasive surgery and has many advantages in terms of beauty. However, medical doctor's fatigue due to repetitive operations such as tissue resection and suturing still remains a problem to be improved. To solve this problem, there are a lot of automation researches on surgical robots [1], [7]-[10]. Especially in cutting automaton, for high accuracy, optimal path planning is essential factor. Probabilistic Roadmap (PRM) is a popular method for path planning. It creates path from static environment to desired point without collision. However, this does not show great performance in a dynamic environment. Reinforcement Learning (RL) shows strong performance in an unspecified probabilistic environment and it is widely applied to robot motion learning because learning data is not needed before [4]. In this paper, we suggest a collision avoidance path planning for automation of surgery robot by using PRM and RL in dynamic situation. We found the collision avoidance path through PRM and RL, and used mapping algorithm of coordination system from pixel to world coordination and transformed the coordination system from cartesian space to joint space using inverse kinematics. Finally, we apply it to the APOLLON laparoscopic surgery robotic system developed by KAIST in V-rep simulator. As a result, we confirmed a possible of collision avoidance path planning for automation of resection task for surgery robot.