Dexterous manipulation of objects with unknown parameters by robot hands

Human hands can perform dexterous manipulation of objects after a small learning phase. In this learning phase we receive knowledge about the objects shape and mass/inertia by use of our visual, tactile and force sensors. Using this knowledge we can determine how to move the object accurately around a prescribed trajectory such that the object is not lost during manipulation. In this paper the authors consider the problem of manipulating an object with unknown mass and inertia by a robotic hand. It is shown by experiments that by use of knowledge about the robot hand dynamics and by incorporating an adaptive control scheme, dexterous manipulation of an object with unknown parameters can be obtained