Slippage Prediction in Microrobotic Fiber Characterization

The grasp assessment is one of the hot topics in robotics. The robotic gripper can be equipped with tactile, force and/or torque sensors to monitor the interaction between the grasped object and the gripper. In this study, a grasp assessment protocol of cellulose-based aerogel fibers inside microgrippers is proposed in a microrobotic platform which is dedicated to tensile testing of short natural fibers. In this study, the positions of micro-actuators and the microscopic images of the grasped fibers are used to predict the suitability of the grasp for tensile testing as an example of a fiber characterization task. Employing conventional machine learning methods and deep neural networks, the slippage of an aerogel fiber inside the microgrippers is predicted before starting the task of tensile test. The visual data are used for training of a convolutional neural network which results in an accuracy of 85% and recall of 68% in average. By using position data of microactuators for training Support Vector Machine (SVM), Logistic Regression, and K-Nearest Neighbor (KNN) models, an accuracy of 72% and recall of 63% are reached in average.