Assistive Robot Manipulator Pose Prediction based on Object Orientation using CNN

Abstract

Robot, as dictated by its assistive function for human, frequently needed to perform object manipulation (i.e. robotic manipulation). A successful robotic manipulation not only determined by the robot actuator, it is also strongly affected by the robot perception in capturing the world and the workpieces. Robotics has made extensive use of camera-based visual perception, especially robotic manipulation. This perception will enable the robot to perform object(workpieces) detection and localization and generate suitable control policy to move the robot via a series of robot poses. This paper proposed an assistive robot manipulator pose prediction based on object orientation using CNN. The manipulator pose prediction will enable further action for a successful robotic manipulation by providing a relative pose between the object and manipulator. Experiment result shown that the presented method was able to predict the robot pose, especially on the fifth joint of the robot accurately based on the object orientation images.