POLICY GRADIENT-BASED INVERSE KINEMATICS REFINEMENT FOR TENDON-DRIVEN SERPENTINE SURGICAL MANIPULATOR

Jie Chen, and Henry Lau

Keywords

Inverse Kinematics, Policy Gradient, Reinforcement Learning, Surgical Manipulator, Tendon-Driven

Abstract

Minimally Invasive Surgery (MIS) has attracted continuous interests over last decade due to its better surgical outcomes than that of conventional open procedures. However, conducting MIS effectively requires long term training and special expertise for the surgeons and physicians, which highlights the importance of robot-assisted MIS. A 2-DoF tendon-driven serpentine surgical manipulator (TSM) is designed in this work requiring only a trocar with radius of 3.5 mm for Single Port Laparoscopy (SPL). Based on Piecewise Constant Curvature Assumption (PCCA), the inverse kinematics (IK) of the system is derived. Then a novel policy gradient algorithm is developed to refine the derived IK to compensate for system internal nonlinearities, of which the performance is evaluated by a number of trajectory tracking tasks. Rooted mean square error (RMSE) of trajectory tracking experiments in real world with the original IK is 9.693 mm. After deploying reinforcement learning, RMSE reduces significantly to 1.101 mm after only 45 iterations. Therefore, the proposed method provides an efficient alternative to enhance the motion control accuracy for the challenging tendon-driven serpentine surgical manipulators, and furthers its application to robot-assisted interventions in MIS.

Important Links:



Go Back