Robotic Arm Guided by Deep Neural Networks and New Knowledge-Based Edge Detector for Pick and Place Applications

Abstract

This paper presents the preliminary results on the robotic handling for industrial products with various geometric shapes. The main contribution of the present study is to two-fold: the first one is to set up a robotic arm for pick-and-place operation, along with all necessary inverse-kinematics and simulation environment setting, and the second one is to propose a new edge detection approach, which produces robust training patterns and also able to precisely determine the center of mass of the objects. The new edge detection relies on the knowledge-based rules, which emphasize the neighboring pixels on the edge. A deep learning classifier is trained using a dataset which consists of the edge information of different shapes of the objects with diverse orientations and illumination. The preliminary results show that the method is helpful in recognizing the objects correctly and does not be affected by illumination and orientation. After successful recognition, the center of the object is being extracted and the information is passed to a micro-controller which guides robotic arm for a pick-and-place operation of the objects.