Learning Based Estimation of 7 DOF Instrument and Grasping Forces on the da Vinci Research Kit

Abstract

Present-day minimally-invasive surgical robots, such as the da Vinci <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">®</sup> , cannot directly sense interaction between the robotic instruments and the patient anatomy. This includes the instrument grasping force and the 6 degree-of-freedom (DOF) force/torque (wrench) between the instrument and the environment. Previous works have investigated model-based or data-driven methods that use available measurements, such as joint positions, velocities and torques, to estimate either the grasping force or the 6 DOF wrench. This paper extends prior work by developing and evaluating a data-driven (learning-based) method to simultaneously estimate the grasping force and external wrench. This task is complicated by the mechanical coupling between the gripper and other wrist joints, but the network is able to simultaneously estimate external forces, torques, and gripper force with RMS errors of 1.4N, 0.04Nm, and 0.1N, respectively. In addition, transfer learning is shown to enable the neural network to quickly adapt to different instruments.