3D Map Generation for Decommissioning Work

At the Fukushima Daiichi Nuclear Power Station (NPS), robots are used to propel the decommissioning work. Creating a 3D map of the internal environment of the decommissioning work is a necessary technology for improving the working efficiency of a decommissioning robot. The purpose of this research is to realize a 3D map creating a system using a camera by operating the humanoid general-purpose robot Pepper using the Robot Operating System (ROS) and implementing and executing Visual SLAM. The system aims to be applied to a decommissioning robot in the future. In this study, we implemented Visual SLAM methods R-tab Map, which is a method for constructing 3D maps in real-time, and Large-Scale Direct SLAM, which is method that generates a map using luminance with a large gradient between frames without using features for map generation. We also compared and evaluated the effectiveness of the generated maps. Besides, the robot was manually operated using nao_teleop, which can operate Pepper from the ROS library with the PS3 controller, the visualization software rviz, and the point cloud visualization library PointCloudViewer. Using two Visual SLAM's methods, we implemented the experiment. During the experiment, a chair was placed in front of Pepper as an obstacle. In this environment, Pepper was rotated 360 [deg] by manual operation. However, nao_teleop and Visual SLAM methods caused a conflict between the manual operation of Pepper and the 3D map generation process, and both processes stopped. Therefore, the 3D map wasn't generated. To resolve the conflict with operation, Pepper was moved by human and the experiment was again. As a result, the camera image obtained from Pepper was distorted, the map could not be optimized Therefore, the part of 3D map was only generated. For this reason, to realize a 3D map generation system, it was important to properly calibrate the camera that could acquire a flat image without distortion.