POSE ESTIMATION USING AUGMENTED REALITY FOR PARTS UNDER ROBOTIC REPAIR

抄録

The combination of additive manufacturing techniques such as cold spray coating and a controlled robot arm can yield a new generation of robotic repairing systems. To improve the efficiency, it is important to rapidly acquire accurate pose information of the part being serviced, which is then used to guide inspection and repair. Previous approaches are labor intense and lack the capability of direct visual examination. In this research we propose to develop 6-DOF pose estimation of the part and the region of the part to be inspected utilizing augmented reality with a HoloLens. It requires only the 3D object (.obj) file, the HoloLens camera, and ArUco markers printed onto a dodecahedron to be viewed at any angle. The pose estimation is solved through an optimization problem, given the coordinates of specific points in model frame coordinates and their location in the image. The location in the image is determined by pose estimating a stylus tip attached to an ArUco marker dodecahedron and saving it with respect to a fixed workspace ArUco marker dodecahedron. This point is reprojected to an image with the workspace dodecahedron in view. We also present an approach to quickly view these pose estimations after the angle of the camera has changed by saving the pose estimation as a transformation matrix with respect to a workspace dodecahedron, and then translating it back to the HoloLens coordinates. The effectiveness and efficiency of the proposed method is demonstrated through an experimental case investigation.