EAGER: Characterizing Physical Interaction in Instrument Manipulations

随着个人机器人在我们的家庭中变得普遍，它们将为主人执行广泛的有益任务。在这样做的过程中，机器人和家庭环境之间会发生各种各样的物理互动，例如，拿起一个马克杯需要机器人在抓住它的时候小心地控制接触力。当任务变得更高级时，例如操作开罐器，机器人将必须“理解”如何抓住罐头和开罐器，以便操作开罐器。该项目的目标是收集数据并开发算法，使机器人能够理解如何掌握工具，从而促进其安全有效地操作工具的能力。本研究旨在提高对日常操作任务中仪器与环境之间物理相互作用的理解，以开发有效的机器人抓取和操作计划来促进任务。本研究设计并开发了一个物理交互观察系统，以观察多个参与者在几个具有代表性的仪器操作任务中仪器与环境之间的交互运动和扳手。每个操作任务都有其仪器运动模型和扳手分布模型。利用该模型，利用任务扳手覆盖度量生成最优抓取，并利用真实机械臂和手平台进行评估。本工作收集的数据使研究人员能够充分探索日常生活任务，并为其他相关研究提供优秀的训练和测试数据集。最优操作抓取方法使机器人能够以牢固的抓取方式握住仪器，以抵御日常生活交互中的干扰，高效地执行任务。

As personal robots become common in our homes, they will perform a broad range of helpful tasks for their owners. In doing so, all sorts of physical interactions will occur between the robot and the home environment, for example, picking up a mug requires the robot to carefully control contact forces as it secures a grasp. As tasks become more advanced, for example, operating a can opener, the robot will have to "understand" how to grasp the can and the can opener, so that it can operate the opener. The goal of this project is to gather data and develop algorithms that will allow a robot to understand how to grasp tools in a way that facilitates its ability to operate that tool safely and effectively. This research aims to improve the understanding of the physical interactions between the instruments and the environment in daily manipulation tasks with the goal of developing effective robotic grasp and manipulation planners to facilitate the tasks. The research designs and develops a physical-interaction observation system to observe both the interactive motion and wrench between the instruments and environment in several representative instrumental manipulation tasks by a number of participants. Each manipulation task is characterized with its instrumental motion models and wrench distribution models. Using the models, optimal grasps are generated using the task wrench coverage measure and evaluated using a real robotic arm and hand platform. The data collected in this work enables researchers to fully explore daily living tasks and provide excellent training and testing data sets for other related research. The optimal manipulation grasping approach equips robot manipulators with the ability to hold instruments with a firm grasp to withstand the disturbance in daily living interactions and perform tasks efficiently.