A Learning-from-Demonstration Based Framework for Robotic Manipulators Sorting Task

This paper presents a safe and effective framework based on Learning from Demonstration (LfD) for robotic manipulators sorting task, which includes six parts: learning a task model, binarization, contour detection, Principal Component Analysis (PCA) training and recognition, grasping objects, experiments. We develop a sorting task for a 7-DOF manipulator. The purpose is to use the manipulators as a co-worker in an industrial environment. Firstly, we implement the Learning from Demonstration (LfD) method to make a better communication between robots and humans. A computer vision algorithm to realize the contour detection for this task is developed subsequently. Finally, we apply machine learning algorithm named PCA to the object recognition area. We illustrate the effectiveness of the proposed framework by performing a sorting task with a Baxter robot.