Authors: Elio Ogas; Luis Avila; Guillermo Larregay; Daniel Moran
Addresses: Laboratorio de Mecatrónica, FICA-UNSL, San Luis, 5730, Argentina ' Laboratorio de Investigación y Desarrollo en Inteligencia Computacional, FCFMyN-UNSL, San Luis, D5700HHW, Argentina ' Laboratorio de Mecatrónica, FICA-UNSL, San Luis, 5730, Argentina ' Laboratorio de Mecatrónica, FICA-UNSL, San Luis, 5730, Argentina
Abstract: The latest advances in human-robot interaction make possible the spread of collaborative robotic systems in the industrial environment. In this context, most of the industrial robots will serve as human assistants in assembly lines, involved in complex manufacturing tasks. This implies the need to improve the ability of robots to manipulate objects in unstructured scenarios that are often prohibitively expensive to model. In this work, we use a convolutional network for recognising different objects on the work plane. Later, a grasping algorithm is used to estimate the best robot gripper posture so that the robot can pick up the desired object from the cluster. We use the Hough transform and the friction cones theory to identify a set of candidate grasping points on the piece. In general, we found that our implementation performs well and the robot was able to pick up a variety of objects.
Keywords: industrial robot; deep learning; object grasping; Hough transform; friction cones.
International Journal of Mechatronics and Automation, 2020 Vol.7 No.3, pp.113 - 121
Received: 15 Feb 2019
Accepted: 27 Feb 2020
Published online: 13 Aug 2020 *