Smart Vision-Guided Robotic Depalletising System
Author
Soós, Christoffer Johan
Term
4. semester
Education
Publication year
2026
Submitted on
2026-01-30
Pages
51
Abstract
Automated depalletising is a massive challenge in warehouse automation due to high variability in parcel size, shape, material, and stacking configuration. While recent advances in AI-based perception have enabled robotic systems to operate in mixedSKU environments, limitations remain in robustness, autonomy, and generalisation under unstructured conditions. This thesis investigates the design and implementation of a robotic depalletising system that integrates vision-based object detection, segmentation, and pose estimation. A modular perception pipeline is developed to process RGB-D data and extract graspable objects from pallet scenes. The system employs depth-based segmentation combined with edge-based detection for accurate box identification, and RANSAC-based plane fitting for pose estimation without requiring prior knowledge of parcel dimensions. The system is implemented within a ROS2-based architecture using a UR10 collaborative robot, VG10 vacuum gripper, and Intel RealSense D455 RGB-D camera. Experimental evaluation across three scenarios of increasing complexity shows 93.3% detection accuracy and 83.3% overall pick-and-place success rate.
Documents