虚拟手自然抓取动作生成研究

It's essential to generate stable grasp postures in virtual hand grasp planning. However, unnatural postures would occur when only stability is considered. To generate natural postures that guarantee a physically natural appearance, this paper presents a grasp planning algorithm of the virtual hand. First, the algorithm builds a mapping between object and natural grasp postures database, which is extracted from human behaviors according to the GRASP taxonomy. Guided by the human grasp posture, a set of approximate natural grasp postures can be found by searching in the configuration space of the hand. Besides, this method segments the complex objects into graspable parts and then plans for natural grasps. We test the algorithm with five typical grasp postures in the database on various 3D objects, and the experimental results show the grasps are more natural than those obtained from the previous methods.