| Robot autonomous grasping technology is a hot spot in the field of intelligent robot research.With the development of robot technology,people have higher and higher requirements for robot autonomous grasping.However most of curent robotic grasping planning methods are based on 2D or 2.5D input information,and most robotic grasping uses rigid grippers.In this paper,we proposes a six-degree-of-freedom grasping planning algorithm for robot soft hand with strong generalization ability.In order to make full use of 3D information of objects in grasping planning,a robot grasping planner in the form of a three-dimensional convolutional neural network will be constructed in this paper.We first establish a six-degree-of-freedom grasping model of the robot soft hand,namely a grasping probability map containing grasping probability,grasping point,grasping direction and wrist rotation angle.After a series of operations such as 3d data format conversion,point cloud filtering,point cloud segmentation,voxelizing point cloud data and etc,the grasping dataset of this subject was made for subsequent training of grasping neural network.In this paper,a three-dimensional convolutional neural network for grasping planning of robotic soft hand is built to achieve the purpose of inputting three-dimensional information of the object to be grasped and outputting corresponding grasp prediction to be calculated by the computer for the motion planning of the robot arm,which is a two-level neural network.Finally,the confirmatory experiments of the robot soft hand autonomous grasping technology are completed in this paper.The grasp planning algorithm of this subject,the trajectory planning algorithm of the 6-degree-of-freedom manipulator,and the communication program between the parts have been experimentally verified,verifying the feasibility of all the algorithms.Results of experiments show the effect of the robot soft hand grasp algorithms of this subject is great. |
PDF Full Text Request