Robotic Collaborative workstation based on Object 6DOF Pose Estimation
Industrial manufacturing assembly lines are characterized by performing manual tasks that lead operators to outstanding workload, both physically and mentally, that end up in stress and injuries. This thesis delves into the implementation of an Industry 5.0 robotic collaborative workstation equipped...
| Author: | |
|---|---|
| Format: | master thesis |
| Publication Date: | 2024 |
| Country: | España |
| Institution: | Universidad Loyola Andalucía |
| Repository: | Brújula |
| OAI Identifier: | oai:repositorio.uloyola.es:20.500.12412/6137 |
| Online Access: | https://hdl.handle.net/20.500.12412/6137 |
| Access Level: | Open access |
| Keyword: | Instance Segmentation Object Pose estimation 3D reconstruction Robotics Path Planning Computer Vision |
| Summary: | Industrial manufacturing assembly lines are characterized by performing manual tasks that lead operators to outstanding workload, both physically and mentally, that end up in stress and injuries. This thesis delves into the implementation of an Industry 5.0 robotic collaborative workstation equipped with cutting-edge Artificial Intelligence (AI) systems for an effective human-robot interaction. The main functionality of this system is the implementation of computer vision algorithms to locate tools and perform robotics gripping. Foundation Pose is the system which provides the position and orientation of unseen objects to the robot, enabling an effective path planning and object gripping task. Accurate segmentation and 3D models are required for performing that pose estimation, employing and evaluating CNOS and SAM-6D systems for the generation of compatible instance segmentation masks, and BundleSDF for reconstructing 3D models of real-world objects. The final stage of this thesis includes the robotics path planning for object manipulation. This research compares the effectiveness of Reinforcement Learning PPO algorithm, contrasting it with traditional robotics methods using Moveit! Library, ending with object gripping. The results obtained are promising, reaching and picking the object successfully in 10 trajectories. This means a perfect synergy between Foundation Pose, CNOS (which resulted better for segmenting), and its transformation to robot coordinate system. This system is already part of CATEC’s pilot factory of the project 5R Network and offers encouraging results that could impact the aerospace industry. Despite further advancements are needed for real-world factory robustness, this research provides valuable insights for integrating this framework into actual aerospace manufacturing environments. |
|---|