In this master's thesis, we addressed topics related to the applications of inserting a vertical load-bearing post into a lattice base and inserting a transverse connecting crossbar into an adjacent vertical post. These topics include the peg-in-hole problem, simultaneous insertion of two pegs into two holes, and convolutional neural networks.
We developed two applications. The application for inserting the vertical load-bearing post into the lattice base represents a peg-in-hole problem. In determining the insertion points, we used the YOLACT neural network, which is designed for object segmentation in images. For insertion, we only needed to determine the insertion axis, which we achieved by aligning the camera with the center of the base and reading the camera coordinates in space. The application for inserting the transverse connecting crossbar into the adjacent vertical load-bearing post presented a more complex problem, which we approached as the simultaneous insertion of two pegs into two holes. For successful insertion, we needed to determine the insertion points in 3D space based on a captured 2D image. This was feasible because the insertion points, due to the design of the lattice structure, are located at predefined heights. We first designed both applications in a simulation environment and then validated the results in a laboratory setup.
|
