Motor Skills Learning for CNT Pick-up of Micro-Nano Robotic Manipulator in SEM

Abstract

The task of picking-up carbon nanotube (CNT) by the atomic force microscope (AFM) cantilever in the scanning electron microscope (SEM) was divided into several meta-tasks in this paper, and the motor skills of the complete task were learned from the simple meta-tasks, so as to learn the motor skills of the micro-nano robotic manipulator. Firstly, according to the motion characteristics and working environment of the manipulator, a segmentation criterion was established to divide the manipulation tasks into several different meta-tasks. Secondly, the motion trajectories of the same meta-tasks divided by multiple demonstration was filtered by moving average filter, and procrustes dynamic time warping (pDTW) was used for timing alignment. Then, Gaussian mixture model (GMM) was used to characterize the motion characteristics of the meta-tasks, and the optimal motion trajectories of the meta-tasks were generated by Gaussian mixture regression (GMR). Finally, according to the optimal meta-task trajectories, Dynamic Movement Primitive (DMP) was used to learn the motor skills of meta-tasks, and the meta-tasks library was recorded and created. We learned the motor skills of micro-nano robotic manipulator working in micro-nano environment, which laid a foundation for autonomous movement to complete tasks in the future.