부산대학교 도서관

This paper proposes an autonomous sensor-based control of an aerial manipulator for tracking horizontal pipe structures with continuous contact. This approach can be used for continuous contact inspection of long-range industrial pipe structures. To automate the aerial manipulator, the perception module of the proposed approach includes a deep learning technique for pipe identification in the image, a classical vision for feature extraction, and a Kalman filter to increase the frequency of data flow in a system. The integration of LiDAR and the camera sensor has been used to locate the approximate contact point. We have designed and developed a novel manipulator along with feed-forward position control to maintain continuous contact with the pipe. The sliding mode controller has been developed for lateral, yaw, and altitude motion control of the aerial platform. The novel forward velocity function is introduced which serves the purpose of smooth tracking along with maintaining a pipe in the reachable space of the manipulator. The fully autonomous operation strategy has been designed to organize subtasks sequentially with feedback. During the tracking of a pipeline, the lateral and altitude position of pipe with respect to the manipulator's base did not deviate beyond ± 0.2m. Fully autonomous sensor-based control of aerial manipulator has been validated experimentally on a 10-meter-long pipeline.