A Study of Graspless Manipulation

在这项研究中，我们提出了“无抓握操作”，这是机器人操作的一个新概念。在人类操作中，你不会牢牢地抓住一个物体。你可以通过推动、翻滚等方式灵巧地移动物体。在这种操作中，一个环境，如桌子或地板，支撑着物体的几乎所有重量。我们将这种灵巧的操作方法应用于机器人的操作。无抓握操作的介绍首先，我们将无抓握操作定义为使用与环境接触的操作。根据与环境的接触状态，将无抓握操作分为推、翻、转三种类型。这三种无抓法各有不同的特点。例如，在移动速度，或，移动分钟距离等。然后，您可以组合三种类型来实现各种操作任务。在无草操作中，存在着“机械手—对象—环境”的闭环连接机制。为了应对这种循环，我们没有采用力控制的方法，而是借鉴人的方法提出了两种基于位置控制的方法。一种是利用自由关节吸收位置误差的方法。另一种是利用刚性手指和环境的几何约束的可移动空间的方法。具有视觉反馈的无抓握操作在无抓握操作中，操作者不刚性地抓住对象。这样就不能根据机械手的位置和姿态来计算物体的位置和姿态，并且由于摩擦的不确定性，物体不能根据估计进行运动。我们用手眼系统来解决这些问题。我们用视觉观察来跟踪物体，并识别物体的行为来对一些事件进行分类。然后我们用这个手眼系统做实验只用一根手指操纵一个物体。

In this research, we propose 'graspless manipulation, ' a new concept of robot manipulation. In human manipulation, you don't grasp an object rigidly. You move the object dexterously by pushing, tumbling, and so on. In this manipulation, an environment, like a table or a floor, supports almost all weight of the object. We apply such a dexterous manipulation to robot manipulation.1.Introduction of graspless manipulationFirst, we define a graspless manipulation as a manipulation using a contact with an environment. Then we show that, in the view of the contact status with environment, graspless manipulation is divided to three types, pushing, tumbling, and pivoting. These three graspless manipulation method have various different characteristics. For example, in movement speed, or, moving minute distance, etc. Then, you can combine three types to achieve various manipulation tasks.2.Graspless manipulation referring to human methodThere exists a closed loop link mechanism, 'manipulator--object--environment' in graspless manipulation. To cope with this loop, we don't use a force control method, but propose two position control based method refering to human method. One is a method using free joint to absorb position error. The other is a method using a movable space of a geometric constraint by rigid fingers and environment.3.Graspless manipulation with visual feedbackIn graspless manipulation, manipulator doesn't grasp an oject rigidly. Then the position and posture of the object can't be calculated from that of the manipulator, and the object can't move acorrding to estimation because of uncertainty of friction. We use a hand-eye system to counter these problems. We use visual observation to tracking the object and to recognize a behavior of the object to classify some events. And then we have experiment with this hand-eye system to manipulate an object only by one finger.