Robot Grasp Learning in a Teleoperated System
ROBOT GRASP LEARNING IN A TELEOPERATED SYSTEM C. FERNANDEZ, N. GARCÍA, J.M. SABATER, L.M. JIMENEZ, J.M. AZORÍN Miguel Hernández University ABSTRACT A learning approach for robot grasping control is presented. The proposed strategy relies on learning from human demonstration. For this purpose, a teleoperated system has been developed so that the user can give examples of successful operations to the system. The application this system is oriented to deals with automatic classification and storage of objects directly from a conveyor belt. The objects to be picked up by the robot show different shapes and sizes and stand over different sides, so there is a big variability in the tests. As the objects have to be picked and then classified, both the grasping points and the classification criteria must be learned from the examples given by the human operator. The learning algorithm is based on decision trees as this is the method which fits best the application purposes, mainly due to its descriptive power compared to other methods like neural networks. KEY WORDS: Robotics, Teleoperation, Grasping, Machine learning, Decision tree.