Study On Grasping Plan For Multi-targets Based On Depth Data

As an important capacity of robots,researches on the robot grasping have been explored for several decades.A lot of results have been achieved and applied into many kinds of industrial productions.However,the skill of service robots in grasping in clutter is still under development,especially in the complex scene that concludes multi-target.It is a difficult problem to pick out the desired objects quickly and accurately.Past works of grasping were concentrated on single object in clutter.However,there are usually more than one object in the domestic environment.In order to improve efficiency of manipulating in the complex scene with multi-target,a grasping plan method base on depth data is proposed for robot.The key issue of grasping which conclude scene segmentation,object recognition and pose estimation,grasping strategy planning,and grasping motion planning are analyzed and discussed,and the solution that could solve the problem is then proposed.The main contents are as follows:(1)The depth data and the 3D point cloud data of the scene is obtained using RGB-D camera.Gradients of the depth image are calculated,and depth discontinuities are then generated by setting a threshold value.The coefficient of table plane is iteratively calculated using RANSAC algorithm,and the RAN SAC discontinuities are generated by extracting the point cloud above the table plane.Overlaying the depth discontinuities over the RANSAC discontinuities to accomplish scene segmentation.(2)The library of model is building by combining KD(K-Dimension)tree and CVFH(the Clustered Viewpoint Feature Histogram)descriptor,the robot recognize objects and estimate pose of objects using the library.The external parameter between NAO Humanoid and RGB-D camera is calibrated,and the grasping pose of the robot is generated from the pose in RGB-D coordinate system according to the external parameter.(3)The maneuverability of targets is analyzed by building actual grasp space and object space.The evaluating value of targets' maneuverability in all grasp sequences and postures are calculated using the proposed grasping evaluation function.A grasp strategy that has the highest score in the grasp evaluation is chosen as the best grasp strategy of the robot in a clutter.The angular displacement of joints is calculating using Inverse kinematics.The robot manipulates all of the targets in a clutter by driving the joints movement according to the best grasp strategy.(4)In order to verifying availability and effectiveness of the proposed method,the approach is applied to a RGB-D sensor(AUSU Xtion Pro Live)and a NAO humanoid robot.