Designing Robotic Desktop Assistants Using Dynamics-based Shape-Sensing

摘要IIS-9820180 Erdmann，MichaelCarnegie Mellon University$63,876-12 MOST。使用基于动力学的形状感知设计机器人桌面助手这是一个持续三年的奖项的第一年资助。这项研究旨在了解机器人感知形状的操作，以提高机器人与环境交互的多功能性。随着机器人变得越来越像人类一样辅助，它们将需要操纵各种不同形状和大小的新奇物体。识别和操纵这些物体的关键将是许多被动和主动感觉模式的整合。本研究以主动触觉操作策略的信息量为研究对象。例如，当抓取已知形状和动力学但未知姿势的对象时，手指上接触点的时间历史通常提供足够的约束来揭示对象的方向。更困难的是拾取未知形状和未知动力学的物体的任务。同样，触摸对象所需的手指配置对对象的可能形状提供了约束。随着手指的移动和对象在手指中的滑动，这些约束会随着时间的推移而演变，从而允许重建对象的形状。这些任务之间的差异主要在于触摸对象所需的手指数量、主动操纵的程度和内部模型的特征。在已知和未知形状的这两个极端之间存在着更典型的任务，其中存在部分形状信息。例如，人类或机器人可以通过肉眼观察知道物体的粗略形状和姿势；抓取策略然后揭示精细的形状和动力学。这项研究最直接的影响在于操控，可能还会对触觉显示器和工业自动化产生额外的影响。这项研究产生的形状感应程序将使机器人编程变得更容易。这项研究不需要机器人程序员精确测量物体的形状和位置，而是提供在机器人操作过程中提取这些信息的自动方法。这种自动化方法将促进机器人辅助。此外，了解操作过程中产生的形状信息的性质将有助于设计更好的触觉设备。

Abstract IIS-9820180Erdmann, MichaelCarnegie Mellon University $63,876 - 12 mos.Designing Robotic Desktop Assistants Using Dynamics-Based Shape-SensingThis is the first year funding of a three year continuing award. This research seeks to understand shape-aware robotic manipulation in order to improve the versatility of robots interacting with their environments. As robots become more human-assistive, they will need to manipulate a large variety of novel objects of varying shapes and sizes. Key to recognizing and manipulating these objects will be the integration of numerous passive and active sensory modalities. This research focuses on the information content of active tactile manipulation strategies. For instance, when grasping an object of known shape and dynamics but unknown pose, the time history of the contact points on the fingers often provides enough constraint to reveal the orientation of the object. More difficult is the task of picking up an object of unknown shape and unknown dynamics. Again, the finger configurations required to touch the object provide constraints on the possible shapes of the object. As the fingers move and the object slips in the fingers, these constraints evolve over time, thereby permitting reconstruction of the shape of the object.The difference between theses tasks lies primarily in the number of fingers required to touch the object, the degree of active manipulation, and the character of internal models. In between these extremes of known and unknown shape lie more typical tasks, in whichpartial shape information exists. For instance, a human or robot may know the rough shape and pose of an object from visual inspection; grasping strategies then reveal the fine shape and dynamics. The impact of this research lies most directly in manipulation, with possible additional impact on haptic displays and industrial automation. The shape-sensing routines produced by this research will make it easier to program robots. Rather than requiring a robot programmer to measure object shapes and locations precisely, thisresearch will provide automatic methods for extracting that information during robot manipulation. Such automatic methods will facilitate robotic aids. Moreover, understanding the nature of the shape information produced during manipulation will facilitate the design of better haptic devices.