虚拟手自然抓取动作生成研究

Research on Natural Grasp Generation of the Virtual Hand

在虚拟抓取规划中,生成稳定的抓取姿态非常重要,但是只考虑稳定性往往会导致抓取姿态不符合自然的人手抓取习惯.为了生成既满足物理约束又符合人手抓取习惯的自然抓取姿态,提出一种基于抓取分类理论的虚拟手自然抓取规划算法.首先根据人类抓取分类理论,基于人类行为构建物体形状与人手抓取姿态之间的映射关系,然后在动作规划中加入人手抓取姿态的引导,最终在虚拟手的位姿空间中找到一组自然的抓取姿态.对于复杂的多部件物体,先将其分割成可抓的部件,再对可抓部件进行自然抓取动作的生成.实验中展示了以人类抓取分类理论中的5种人手抓取姿态为指导,对常见的三维物体生成自然抓取姿态的结果,经用户评估,所提算法生成的抓取动作相比其他算法更符合人手抓取习惯.

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.