AUTOMATIC AND OPTIMIZED VIEWPOINT PLANNING FOR OBSERVATION TRAJECTORIES IN ROBOTIC INSPECTION

Industrial automation plays a key role in improving the efficiency and reliability of production processes, particularly in the context of quality control. A highly effective approach involves the use of robotic systems equipped with 3D scanners, capable of capturing point clouds for comparison with known reference models. Viewpoint planning - deciding the pose to observe an object - is critical to ensure complete and accurate data acquisition, and manually programming this scanning routine turns out to be mostly a time-consuming and tedious process. Then, its automation becomes a crucial step for improving the scanning task. This thesis, developed in collaboration with Euclid Labs S.R.L., focuses on the development of an adaptive and automated view planning strategy that allows our 3D vision system, composed by a robotic manipulator with an eye-in-hand camera, to perform the object inspection efficiently. The work is conducted in a simulated environment and developed in C\# programming language using the Visual Studio IDE. The core of the implemented algorithm is based on the use of Mass Vectors Chains, which guide the automatic and adaptive selection of viewpoints based on information obtained from previous scans and determining the Next-Best-View by giving greater weight to the object patches with the lowest percentage of scanned area. Through the analysis of point clouds acquired from different simulated positions, the aim is to optimize the coverage and quality of the acquired data while reducing scan times. The validation of the proposed method is carried out by modifying key working parameters within the simulation environment and by comparing the MVC approach with its enhanced version, which integrates K-means clustering. This allows for a comparative evaluation of the effectiveness, efficiency, and robustness of the two configurations under different operating conditions.