Dynamic guidance virtual fixture for hydraulic manipulator via learning from demonstration

Abstract

Operating heavy-duty hydraulic manipulators with a master-slave control system is very challenging to execute complex tasks in unstructured environments. In this paper, to improve the operational efficiency for performing repetitive tasks, we designed a dynamical guidance virtual fixture via learning from demonstration to assist the operation. A data fitting method is proposed by reconstituting the control vertexes to address the operation noises caused by response lag and oscillation tendency of hydraulic manipulators, such that a smooth nominal trajectory is obtained and used to generate the virtual fixture. Then, a pipe-constraint guidance virtual fixture is designed with multiple control modes to meet the demands of free and constraint motion. Comparative tests were carried out with a 3-DOF hydraulic manipulator to perform trajectory tracking tasks and movement within a limited space. Compared with no assistance, the results show that the average time of task completion can be reduced by over 50% with the proposed guidance virtual fixture. Besides, the mental pressure of the operator can be reduced since collision avoidance can be easily achieved.