6D Grasping Based On Lateral Curvatures and Geometric Primitives

• Daniel M. de Oliveira UFBA
• Caio C. B. Viturino UFBA
• André G. S. Conceição UFBA

Resumo

The following paper presents a 6D grasping system using a deep learning algorithm to detect and estimate the object pose in 6D in an RGB image paired with a grasping algorithm in the point cloud to estimate the best position to grasp the object. The proposed grasping algorithm uses geometric primitive and lateral curvatures to estimate the best region to grasp an object. The validation of the system was carried out in the Webots simulation environment, using a UR5 arm, Robotiq 2F-140 gripper and an RGB-D sensor. The 3D models used for the objects were made available by the dataset used in the deep learning algorithm.

Palavras-chave: Deep learning, Solid modeling, Three-dimensional displays, Neural networks, Pose estimation, Grasping, Object detection