Grip Force Perception Based on dAENN for Minimally Invasive Surgery Robot

Abstract

Although robot assisted minimally invasive surgery brings the gospel to patients, force perception is gone. Among all the contact forces during surgery, the instrument grip force plays the most important role. In this paper, a grip force perception method based on denoising AutoEncoder Neural Network (dAENN) is proposed. The method utilizes sensor data including encoder readings and motor current over a time window as the input of dAENN for sufficient information. An Artificial Neural Network (ANN) is then introduced as a machine learning tool to learn the nonlinear mapping between the compact features and grip force labels. Feature extraction is first introduced into grip force perception problem in this paper. Experiment results shows adequate expressive capability of the extracted coding as well as the superior grip force perception performance over several popular data-based methods under the same dataset.