Dynamic Movement Primitives for Human Robot interaction
Dynamic Movement Primitives for Human Robot interaction
复制标题
人机交互的动态运动原语
DOI：
--
复制
发表时间：
2012
期刊：
影响因子：
--
通讯作者：
Anthony Remazeilles
中科院分区：
文献类型：
--
作者：
Miguel Prada;Anthony Remazeilles
在这篇文章中提出了一个通用的动态运动原语（DynamicMovementPrimitives）框架的专门化，以正确地解决人类机器人协作的一个关键活动，即对象交换。作为实现这一具有挑战性的技能的第一步，本文侧重于手臂运动，以达到最初未知的交流网站。提出了与该应用相关的两个改进。首先，描述了对所述训练的两个主要分量（分别提供技能形状吸引子和目标吸引子）之间的过渡的更好控制，使得能够定义这两个分量之间的过渡何时以及如何发生。然后提出了一种扩展来处理目标位置沿着时间变化的情况，这提高了轨迹向移动目标（即人类伙伴的手）的收敛。这两个改进是通过比较所获得的行为与通过运动捕捉实现的人类观察进行验证。
A specialization of the generic Dynamic Movement Primitives (DMP) framework is proposed in this article to correctly address a key activity for human robot collaboration that is object exchange. As a first step towards implementing this challenging skill, this paper focuses on the arm motion to reach the initially unknown exchange site. Two improvements related with this application are proposed. First of all a better control of the transition in between the two mains components of the DMP –respectively providing a skill shape-attractor and a goal-attractor– is described, enabling to define when and how the transition in between these two components occur. Then an extension to handle situations where the goal position varies along time is proposed, which improves the convergence of the trajectory towards a moving target (i.e. the human partner’s hand). These two improvements are validated by comparing the obtained behavior with human observations realized through motion capture.