Examenspresentation Aishwarya Kanchan och Yiyun Xia

Examiner: Yasemin Bekiroglu

Opponents: Bernardo Taveira and Ernesto Lozano

Abstract:
The integration of collaborative robots in industrial settings have increased significantly in the
past decade. The integration improves speed, reliability and efficiency of picking tasks.
However, one of the common challenges in bin picking applications revolves around accurately
predicting and executing stable grasps. To address this limitation, this thesis proposes a solution
in the form of a deep learning model trained on singulated industrial objects, leveraging input
features to learn and identify patterns for precise grasp predictions. Grasps are generated from
perceived point clouds, and the model's ability to classify good grasps accurately is thoroughly
analyzed. The communication framework between sensors, the UR10 robot, and users is facilitated
by the Robot Operating System (ROS). The approach of this thesis focuses on designing a grasp
prediction model that exhibits high accuracy, with a comparative analysis against the Grasp Pose
Detection (GPD) method serving as a baseline. The investigation aims to develop a comprehensive
understanding of the factors contributing to generation of successful grasps and identifies required input
features to enhance prediction accuracy.

Welcome!

Aishwarya, Yiyun and Yasemin