Vision-based estimation of excavator manipulator pose for automated grading control

Abstract

Abstract As an effective alternative to manual operation of hydraulic excavators, automated systems are being developed for common tasks that excavators routinely perform on typical job sites. An example is an automated ground grading system that can enhance the productivity of excavators by assisting the operator to perform ground grading in a fast and accurate manner. To develop an automated grading system, sensors are needed to measure the position of the hydraulic manipulator linkages for the feedback control system. In this paper, a novel method to estimate the pose of a hydraulic manipulator using a vision-based neural network system is presented. A webcam is used to capture images of a moving manipulator, and the captured images are used to train a neural network. Then, the trained neural network can be used to estimate the pose of the excavator manipulator for the feedback control system. A simulation study shows a stable grading performance when a PI controller is used to control the manipulator based on the estimated manipulator pose.