3D Pose Estimation for Robotic Grasping Using Deep Convolution Neural Network

摘要

With the progress of artificial intelligence, robots begin to enter family service. Autonomous object grasping in a cluttered scene is the most frequent operation of a service robot in daily life while it is still a challenging problem in the field of robotics. In this paper, we develop a robot system using a deep convolution neural network for 3D object grasping. The system is composed by a color camera, and a robot arm with a gripper. The color camera provides robotic vision about surrounding environments; the deep neural network performs an end-to-end mapping from vision images to the 3D poses of the object of interest; the robot arm with the gripper is then driven to grasp the object. In addition, we also present an automatic data labeling method for the training of the convolution neural network. Preliminary experiments were performed to evaluate our robot system and the results have confirmed its effectiveness.