Deep Learning Implementation in Multi-Fingered Manipulator Robot for Pick and Place Food Serving Equipment

Abstract

Travel, tourism and hospitality companies have started to adopt RAISA systems in the form of chatbots, delivery robots, autonomous dishwashers, conveyor restaurants, self-service information kiosks and many others [1], [2]. This research focuses on the implementation of deep learning artificial neural networks for object recognition in determining the pose estimation of the manipulator robot and planning the grip on the end effector. A robotic manipulator with 4 degrees of freedom is used to support the estimation of pose angles and an end effector in the form of a 5-finger gripper is used to obtain various grips on objects with random shapes. An RGB camera is used for object recognition with an eye-on-hand configuration, which is linked to the end effector to obtain visual information on objects using the YOLOv3 deep learning algorithm. The end effector works optimally on objects with the basic shape of a tube, rectangular prism, hexagon prism and ten-sided prism with a maximum load that can be lifted of 303 grams with a success rate of 71.23%.