차압 및 심층신경망 기반 유압 로봇팔 끝단 반력 추정

Abstract

In this study, a method to represent reactive forces at a stick-type controller has been proposed using a haptic master device to effectively communicate work status to users during subsea fracture operations through a teleoperated robot. However, estimating reactive forces acting on the tool underwater presents significant challenges. Therefore, a method to address these issues has been developed here that combines differential pressure measurements with a deep neural network (DNN) to estimate the reactive forces at the hydraulic manipulator's tool with good accuracy and a high sampling rate. Specifically, the reactive force was predicted from high-sampling-rate differential pressure data, and the DNN was used to update the reactive force estimation with high accuracy. These tasks were performed recursively within a Kalman filter framework. Furthermore, a plaster fracture experiment was conducted in a terrestrial environment to verify the proposed method. The estimated reactive forces were compared with those measured by a force-torque sensor using data retrieved from the inertial sensors, joint encoders, and other relevant sensors. The differential pressure-DNN-based approach demonstrated high accuracy in estimating reactive forces in key directions while maintaining fast sampling speeds.