Research on robot grasping detection method based on point cloud
This paper proposes a method for detecting the grasping pose of unknown objects in unstructured scenes.Firstly,the initial grasping points are obtained by uniformly sampling the point cloud of the scene,and the local coordinate system of each grasping point is established based on the PCA method.Then,the grasping pose is determined according to the parame-ters of the two-finger gripper,and the candidate grasping poses are expanded using a pose search strategy.Next,the point cloud of the gripper closing area corresponding to each grasping pose is extracted,and the candidate poses are filtered through a grasp quality evaluation network to output poses with higher scores.Finally,a robot grasping simulation platform is built on ros platform for experimental verification.Experimental results show that the proposed method can generate stable grasping poses on the point cloud of the target object,and can complete grasping tasks of single or multiple randomly placed unknown objects in a simulated environment.
unstructureddetecting the grasping posequality evaluation networksimulation
NETL
NSTL
万方数据
维普
