Improvement and Control Trajectory Tracking of a Three-Axis Manipulator for New Training Strategies

Abstract This study discusses the development of training strategies for RRR (3 revolute joints) triaxial manipulators. Two robot structures and a control phantom were constructed using incremental technology. An universal algorithm was developed to process and autonomously repeat the trajectories of robot movements by imitation and learning. The coordinates of the model articulated positions were then saved as coordinates of the manipulator position. For validation purposes, the newly developed training strategies and the repeatability of robot movements were tested in stages. After the first stage of testing in terms of positioning accuracy, structural changes were introduced in the robot by mechanical engineering to improve its manipulation quality. Next, another training strategy for the improved robot version was developed for the same movement algorithm. This demonstrates the applicability of the developed control system in applications with different structures and requirement for high control-command quality (possibility of using the developed algorithm in various robot designs). Experimental results showed that the function of the developed RRR manipulator algorithm enabling the imitation of phantom trajectories and their learning and memorization, along with the consideration of individual training strategies, can be applied to differing structures, achieving positioning accuracy comparable to that of high-class motion equipment, and facilitating navigation of individual robot members. The system composed of the simplified physical model and the manipulator can directly simplify and globalize the control of the robot over significant distances. Additionally, the positioning accuracy was tested using the photogrammetric method, which is a complete novelty in robot positioning research. Furthermore, it was demonstrated that for the first construction of the robot, it is possible to achieve an accuracy of ±1 mm in the automatic replay mode. Moreover, a construction accuracy of ±0.52 mm at a maximum speed of 0.025 s/1° can also be accomplished after manual repeat of the different model movements, i.e., after a learning procedure.