|dc.description.abstract||The amount online purchases conducted by consumers show no sign of declining, and the growth in retail sales have rapidly increased over the last decade. This growth is driving the demand for automated packing and shipping systems, that can perform the highly repetitive work of locating and transporting products from inventory bins to shipping bins. This master thesis addresses this possibility thought the development of an automated bin picking system.
The system was developed at the Agilus robot cell at the Department of Mechanical and Industrial Engineering workshop during the spring semester of 2018. The proposed system utilises a combination of computer vision modules in 2D and 3D, robotics, and deep neural networks. Two neural networks have been trained for the task of performing object detection on specific objects, and evaluated in terms of performance. The kinematic relation between the camera and the robotic manipulators of the cell have been described, and a grasping approach based on a deep neural network has been implemented. A fully automated bin picking system has been realised in the robotic cell at the institute workshop.
Each module of the final system is presented with the associated results. The individual modules are combined to one single system, that is capable of performing autonomous bin picking operations. The final results demonstrate that the system is able to perform the intended task with a relatively high success rate. The results from the individual modules are also presented, and a discussion is presented based on the obtained results. The results obtained from our experiments during practical work are discussed in context to the possibility of implementing a similar system in a real storage setting, where new products are added to the inventory list continuously. Finally, strengths and weaknesses of the proposed system are discussed, along with potential improvements for future work.||