Print Email Facebook Twitter Single-Query Motion Planning for Grasp Execution Title Single-Query Motion Planning for Grasp Execution Author Meijer, J.G.J. Contributor Wisse, M. (mentor) Faculty Mechanical, Maritime and Materials Engineering Department BioMechanical Design Date 2017-05-30 Abstract The grasping of objects is a highly desired function for service robots. To aid with grasping, researchers have developed grasping approaches. A demerit of existing approaches is that they solely focus on the grasp finding. A more important part in grasping is the grasp execution, which involves the solving of a motion planning problem. Currently, 23 sampling-based motion planners can be chosen from the Open Motion Planning Library (OMPL) within MoveIt!, a ROS framework that provides the tools for motion planning. However, no recommendations for selecting a specific planner for high performance is given. Moreover, difficulties in real-world grasp executions are typically not outlined in existing grasping approaches. In this thesis high-performing planners are selected for grasp executions and difficulties in performing such executions in a real-world setup are outlined. The performance of the planners was analyzed by means of solved runs, computing time and path length. Three lightweight manipulators with different characteristics have been chosen to collect reliable data on planner performance. Various grasp executions have been defined with individual goals. One grasp execution incorporated a motion constraint that demands a specific orientation of the gripper. To achieve maximum performance, the parameters of the planners have been optimized. For a grasp execution which starts moving in a confined space towards an open space, high performance was found with mono-directional tree-based planners with goal bias, such as EST, ProjEST, KPIECE and STRIDE. For a grasp execution which starts moving from an open space towards a confined space, bi-directional planners with lazy collision-checking (SBL and LBKPIECE) yielded highest performance. In the investigated grasp execution that incorporates a motion constraint, high performance was obtained with the BiEST planner. The experimental setup for real-world grasp execution was realized with modified ROS packages. Difficulties in real-world grasp executions exist due to inaccurate depth data, small errors in extrinsic calibration and the lack of verification of a grasp execution using an automatic command. Subject Motion planningOMPLROSgraspinggrasp executionMoveit! To reference this document use: http://resolver.tudelft.nl/uuid:8f27f108-9d0b-4355-bf1d-840e4c13ed5e Embargo date 2019-05-20 Part of collection Student theses Document type master thesis Rights (c) 2017 Meijer, J.G.J. Files PDF Thesis_JonathanMeijer.pdf 39.77 MB Close viewer /islandora/object/uuid:8f27f108-9d0b-4355-bf1d-840e4c13ed5e/datastream/OBJ/view