A Deep Learning Approach to Intrinsic Force Sensing on the da Vinci Surgical Robot

Abstract

In robot-assisted minimally-invasive surgery (RAMIS), force estimation remains a challenging issue. We seek to estimate external forces based on available measurements from the joint encoders and motor currents. To this end, we propose a deep learning approach for end-to-end force estimation on the da Vinci Surgical System that is trained using data collected by both moving an instrument in free space and by palpating a tissue phantom that has an embedded force sensor for ground truth. The trained neural network provides reasonable force estimates (within about 1N to 2N precision given a full range of 10N) and is generalizable to other regions of the robot workspace. We further show that our proposed system can provide useful haptic feedback in a pilot study to differentiate stiffness in various tissue phantoms.