期刊名称:International Journal of Advanced Robotic Systems
印刷版ISSN:1729-8806
电子版ISSN:1729-8814
出版年度:2015
卷号:12
期号:11
页码:167
DOI:10.5772/61750
语种:English
出版社:SAGE Publications
摘要:A Gaussian Mixture Model (GMM)-based grasp-pose generation method is proposed in this paper. Through offline training, the GMM is set up and used to depict the distribution of the robot's reachable orientations. By dividing the robot's workspace into small 3D voxels and training the GMM for each voxel, a look-up table covering all the workspace is built with the x, y and z positions as the index and the GMM as the entry. Through the definition of Task Space Regions (TSR), an object's feasible grasp poses are expressed as a continuous region. With the GMM, grasp poses can be preferentially sampled from regions with high reachability probabilities in the online grasp-planning stage. The GMM can also be used as a preliminary judgement of a grasp pose's reachability. Experiments on both a simulated and a real robot show the superiority of our method over the existing method.