Analysis and Grasp Strategy Modeling for Underactuated Multi-fingered Robot Hand Shuangji Yao Qiang Zhan School of Automation Science and Electrical Engineering Robotics Institute Beihang University Beihang University Beijing P.R.China Beijing P.R.China buaayaoshuangji@163.com qzhan@buaa.edu.cn Marco Ceccarelli and Giuseppe Carbone Zhen Lu Laboratory of Robotics and Mechatronics School of Automation Science and Electrical Engineering University of Cassino Beihang University Cassino (Fr), Italy Beijing P.R.China ceccarelli@unicas.it zhenluh@buaa.edu.cn Abstract - A survey for grasping synthesis method with dexterous robot hand is presented in this paper. The difference of grasping characters is introduced between dexterous hand and underactuated hand. Especially the feature of self-adaptive enveloping grasp by underactuated finger mechanism is outlined as having good performance in grasping unknown objects. In order to generate valid grasps for unknown target objects and apply in real-time control system for underactuated robot hand, a grasping strategy for universal grasp tasks is proposed as based on human knowledge analysis. It is composed by off-line neural networks training section and on-line compute section. Firstly, daily grasped objects are used to build a sample space from human experience. Then, the discrete sample space is computed by a fuzzy clustering method. Finally, the data are used to generate grasp decision scheme by rough set mixed artificial neural networks. The choices of grasp configurations for the underactuated robot hand are simulated for with the aim to show the practical feasibility of the proposed modeling method. Index Terms - Underactuated robot hand; grasping strategy modeling; rough set mixed neural network I. INTRODUCTION Multi-fingered robotic dexterous hands have been proposed since 70s of last century, and the design, analysis, and control of such hands have become an active area of research. Many studies focus on the issues of strategy and planning for target objects grasp. It refers to some aspects such as grasping mode choice, grasping position planning for fingers and palm, contacting point selecting, kinematic and torque computation for each joint, operation and stability. Research works are undergoing from different views in order to analyze grasp planning and establish grasp strategy models. Geometry method is an initial solution to build grasp strategy model which is based on the theory of form-closure and force-closure. Nguyen presented a simple algorithm in [1] for directly constructing force-closure grasps. An efficient algorithm for synthesizing grasp is reported in [2] for bounded polyhedral/polygonal objects. An algorithm for computing all locations of frictionless grasping points is proposed in [3]. An improved approach is reported in [4] by using a ray-shooting algorithm to test force-closure for 3D frictional grasps. Optimum functions can also be used to analyze grasp synthesis for multi-fingered robot hand. A task-oriented quality measure is proposed in [5] for evaluating grasp by computing the minimum singular value for a grasping matrix. Stable grasp and form-closure optimum problem are formulated and solved in [6]. General optimality criteria considering the total finger force and the maximum finger force are introduced and discussed in [7]. A general algorithm composed by two computing phases is presented in [8] for optimum dynamic force distribution in multi-fingered grasping. However, it is complex to provide a formulation and a quantification of an evaluation function in a multi-contact grasp system. In addition, the optimality iterative process also needs computation efforts to converge, thus it is difficult to apply in a real time control system. II. GRASP STRATEGY BASED ON HUMAN EXPERIENCE A human hand can grasp one object with different types of configurations. This would lead to different grasping stability and dexterity. The rule of grasp configuration choice for a human is based on daily experience or intelligence consideration before grasp action. Simplifications and assumptions are proposed in [9], which are applied in manufacturing environment for grasp tasks of robotic hands. A knowledge-based approach of robotic hand is described in [10] for grasping unknown objects. A grasp synthesis algorithm is introduced in [11], in which an expert system can generalize application instances by many prepared grasp prototypes. The relationship among grasp task, object geometry, and grasp choice are analysed and reported in [12]. Nevertheless, a grasping strategy model can be used to establish the relationship between target objects and grasp configuration choice. The grasp synthesis as based on human experience can be developed when a target object’s characteristics are known in terms of shape, size, weight, and grasp task. This will make possible to apply in a real-time control system with three necessary characters: 2817 978-1-4244-2693-5/09/$25.00 ©2009 IEEE Proceedings of the 2009 IEEE International Conference on Mechatronics and Automation August 9 - 12, Changchun, China