Example-based grasp adaptation

Abstract

Finding a way to provide intelligent humanoid robots with the ability to grasp objects has been a question of great interest. Most approaches, however, assume that objects are composed of primitive shapes such as box, sphere, and cylinder. In the thesis, we explore an efficient and robust method to decide grasps given new objects that are irregularly-shaped (3D polygon meshes). To solve the problem, we use an example-based approach. We first find grasps for objects geometrically similar to those the system has seen before. For example, if the system has been shown a cup being grasped by the handle, it should now be able to grasp any new cup. There are two problems to be solved in order to adapt example grasps to the new object. First, the system should be able to retrieve objects that are geometrically similar to the given object from the database storing previously seen objects. After collecting objects the system knows how to grasp, it needs to adapt example grasps to new object.Already, there are some working algorithms for the first problem (shape retrieval). Therefore, our main contribution is to present an algorithm that performs grasp adaptation. Before we adapt a grasp, we first find the geometric correspondence between a demo object and new object using probabilistic graphical model. Based on correlation information together with the demo grasp, we generate a grasp for the new object. To ensure that a robot can effectively grasp the object, we adjust the position of grasp contacts until the quality of the grasp is reasonably high. In test cases, the system successfully uses this method to find the correspondence between objects and adapt demo grasps.