Multi-autonomous Robot Sorting System Using Deep Convolution Neural Network

Abstract

This paper presents a multi-autonomous robot grasping system using a novel deep convolution neural network to perform real-time robotic sorting tasks. The system is composed by a color camera, a sorting line and two robotic arms with grippers. A deep convolution neural network is modified to estimate the position and orientation of the targets in the images and output the corresponding image features of the targets. A task assignment system is considered to transmit visual information about the detected targets to robotic arms for grasping, and continue to track the targets until they are no longer visible to the camera. This paper describes how to efficiently distribute sorting tasks without missing or repeating. We then propose an efficiently and semi-automatic method to expand the published dataset for training. Preliminary experiments were performed to evaluate our robot system and the results had confirmed its effectiveness.