許陳鑑Hsu, Chen-Chien游鈞凱You, Jiun-Kai2022-06-089999-12-312022-06-082021https://etds.lib.ntnu.edu.tw/thesis/detail/8c6bbe784368478f32f1668892fe1d1d/http://rportal.lib.ntnu.edu.tw/handle/20.500.12235/116955none['Thanks to the recent advancement in technology, robotic development has been gradually evolved from traditional industrial robots capable of performing repetitive and routine tasks to intelligent robots with autonomous vision-based grasping strategy to satisfy the needs for various applications in industries, for example, flexible production in low-volume automation for small and medium-sized businesses (SMBs). To accurately execute tasks for various robots, including industrial, service, andcooperative robots, robotic grasping of objects dominates the performance and reliability for a robotic system. Therefore, an optimal robotic grasping strategy incorporating a multiple-object pose estimation mechanism is presented in this thesis, where a RGB camera, rather than a RGB-D camera, is only used to estimate the 6DoF object pose in the images through a proposed object pose estimation system incorporating a projection loss function and refinement network. On the basis of the obtained 6-DoF object pose estimation, we can analyze the transformed 3D cloud of the object via the estimated object pose to derive an optimal grasp pose for the robot arm. As a result, the robotic system can efficiently grasp objects with arbitrary poses in the environment.']noneobject pose estimationLINEMODOcclusion LINEMODgrasp strategy學術論文