Grasp synthesis of multifingered robotic hands

Research and development leading to the realization of a fully autonomous and robust multifingered robotic hand have been going on for three decades. Yet none can be found in an industrial application. The difficulties stem from many factors, one of which is that we lack general and efficient algorithms for the grasp synthesis. The grasp synthesis refers to the following problem: given the model of a rigid object, determine a set of finger positions on its boundary that supports the object and makes it resistant to external forces/torques. Such a grasp is also termed as a form-closure grasp and is an essential part of the robot hand control scheme. This thesis makes a thorough study on the grasp analysis, synthesis and related topics, with the particular attention paid to apply the grasp synthesis algorithms to the experimental regrasping task of a rectangular object. In addition, noting the similarity between robotic grasping and fixture layout design in manufacturing, we further apply the grasp synthesis algorithms to the fixture synthesis of generic workpieces. The grasp analysis associates with inquiring whether the given grasp characterized by contact locations and associated contact models is form-closure. By fully exploring the geometric meanings of form-closure conditions, we show that the grasp analysis can be transformed to a ray-shooting problem, which is dual to a Linear Programming (LP) problem based on the duality between convex hulls and convex polytopes. The algorithm developed examines both the frictionless and frictional grasps in two or three dimensions. For the grasp synthesis, we first consider synthesizing a grasp when some of the fingers are located in the known positions on the object, where the possible contact regions and the most favorable positions for the other fingers are calculated through inequality theory and nonlinear optimization techniques, respectively. Then we consider the grasp synthesis from a random initial grasp, where an efficient search strategy is developed and formulated as a quadratic programming problem that leads the initial non-form-closure grasp towards a form-closure configuration. The two cases cover most circumstances in practice and the approaches can be applied to synthesize grasps for both polyhedral and curved objects. The hand kinematics is further taken into account for practicality when the grasp synthesis of a curved object is dealt with. A large number of various numerical simulations verify the efficiency and usefulness of these approaches. Finally, an experiment setup is built on our Yaskawa five-fingered hand system, where the task of rotating a rectangular object through regrasping is completed. The grasp analysis and synthesis algorithms are applied to obtain proper contact locations during the process and grasping forces are optimized with the emphasis on the balance of the internal forces.