Application of simultaneous localization and mapping for large-scale manipulators in unknown environments

Abstract

In this paper, we study the application of simultaneous localization and mapping (SLAM) for estimating the tool center point (TCP) 6 degrees-of-freedom (DOF) pose of a large-scale hydraulic manipulator without a priori knowledge of the environment. We attach a stereo camera near the TCP of the manipulator and perform SLAM by utilizing the open source version of ORB-SLAM2. In offline experiments, the camera frame and the TCP frame are extrinsically calibrated using an iterative closest point search to match a point cloud of poses from the SLAM module with a point cloud of ground-truth TCP poses, which are obtained from joint encoder measurements along with a kinematic model of the manipulator. The estimated TCP trajectory provided by the SLAM is then compared to the ground-truth TCP trajectory. These preliminary experiments show that a pure visual SLAM algorithm can perform reasonably well in this application scenario. Limitations and future work are also discussed.