At Tajfun Planina d.o.o., we are exploring the possibility of robotizing the CNC machining area. In this specific cell, the workers and the robot would work alternately, which brings the risk of foreign objects in the cell. The goal of this master thesis was to develop an application that would scan the robot cell with a depth camera before the robot's shift and detect wheter any object is in the path of the robot's trajectory. Initially, we reviewed existing solutions in the literature. We reviewed and described the robot vision concepts that were used during this work. We also looked in more detail at robot simulators and how they detect collisions. The mathematical basis for collision detection of objects in simulation is presented. Robotic vision was implemented using a combination of the Yaskawa Motoman GP180 robot and the Intel Realsense D435 camera. The robot used the depth camera to scan the space around it, followed by merging and processing the 3D points of the space using algorithms such as RANSAC, alpha shape reconstruction algorithm etc. The main processing was carried out using the Open3D library. The algorithm concluded with collision detection, which was performed using the PyBullet simulator. Finally, a simple user interface was prepared in React. The functionality of the application was tested in several possible scenarios. Our developed algorithm proved to work as expected. The results of the detection and recognition were evaluated as satisfactory. The biggest challenges were posed by the error of the depth camera used, which was reflected in inaccuracies of the collision detection.
|
