Recognition and Removal of Interior Facilities by Vision-Based Robot System

Abstract

For future dismantling jobs in the renovation of the interiors of office buildings, we propose a robotic dismantling system that will assist human workers with the hard works. As an application of the robotic system, this paper presents the process of removing ceiling fixtures, such as Lamp Panels (LPs) and Air Conditioning Vents (ACVs), by man and robot. In this collaboration, a robot arm provides assistance by holding and collecting the fixtures, and the human worker only removes screws and/or nuts. In order to lead the robot to a holding position, the human worker indicates a position on the fixture to the robot with brief and simple instructions. The robot estimates the pose of the fixture through 3D model-based object recognition with a hand-mounted stereo camera. The integration of multiple viewpoints for the robot with an active lighting system enhances the recognition performance against both natural lighting changes at the site and the variability in the pose between the camera and the object to be recognized. As a verification experiment, the sequential removal of several different ceiling fixtures is presented. In this experiment, robust recognition is achieved with an average accuracy of 10 mm. The feasibility of the system is verified by using the completion time and the precision requirements in a practical environment.