Object gripping algorithm for robotic assistance by means of deep learning

Abstract

This paper exposes the use of recent deep learning techniques in the state of the art, little addressed in robotic applications, where a new algorithm based on Faster R-CNN and CNN regression is exposed. The machine vision systems implemented, tend to require multiple stages to locate an object and allow a robot to take it, increasing the noise in the system and the processing times. The convolutional networks based on regions allow one to solve this problem, it is used for it two convolutional architectures, one for classification and location of three types of objects and one to determine the grip angle for a robotic gripper. Under the establish virtual environment, the grip algorithm works up to 5 frames per second with a 100% object classification, and with the implementation of the Faster R-CNN, it allows obtain 100% accuracy in the classifications of the test database, and over a 97% of average precision locating the generated boxes in each element, gripping successfully the objects.