Sensor-based manipulation for multifingered robotic hand.

摘要

This dissertation addresses several important issues of multifingered manipulation. First, a unified Control System Architecture for Multifingered Manipulation (CoSAM2) is proposed. Then, capacitive tactile sensors are developed to measure contact information. Finally, from differential geometry point of view, Cartesian stiffness of multifingered robotic hand is addressed.; CoSAM2 can achieve simultaneously several objectives of multifingered manipulation including: (a) motion trajectory tracking of a grasped object; (b) improving the grasp configuration in the course of fine manipulation; and (c) optimizing grasping forces to enforce contact constraints and compensate for external object wrenches.; Given an initial and a desired final configuration of the grasped object, CoSAM2 generates the trajectory between the two configurations with a desired object velocity. Then, a kinematics based approach is used to plan motion of fingers from the desired object velocity. As this approach does not keep the grasp in optimal position, failure of manipulation could possibly result. A tactile sensor based manipulation approach is proposed. This approach uses grasp constraints to determine motion of the contact points so that the manipulation task can be achieved and the grasp configuration can be optimized. In addition, with the process of manipulation and changing of grasp configuration, finger grasping force have also to be planned and can be optimized in real time to keep the grasp in stability. A nonlinear real time grasping force optimization algorithm has been extended to multifingered manipulation with rolling constraints. The grasp map is updated in real time using tactile feedback.; In the course of multifingered manipulation, the local geometric parameters of the fingertips and object have to be updated in real time. Capacitive tactile sensors are developed to measure the contact information including contact location, contact force and curvature information. Each tactile sensor consists of a 16 x 16 capacitor array. Each tactile element is formed by an upper copper electrode and a bottom copper electrode with an elastic dielectric rubber separation. Once a force exerted on the surface of the sensor, the deflection will occur on the elastic rubber leading to change of capacitance. Measuring the capacitance change and the index of the tactel, contact location and contact force can be obtained. Furthermore, the principal curvature directions of the object can be obtained. From Montana's equations, the geometric parameters can be obtained by experiments. How to detect the local curvature information at the contact point of the object is presented. Contour following of an unknown object is addressed.; Cartesian stiffness is a geometric map which transforms a differential displacement (twist) of the robot end-effector into an incremental change of force (wrench) exerted in the end-effector. The definition of Cartesian stiffness depends on connection. Chosen a symmetric connection ∇ _XY = ½[X, Y] which is compatible with any bi-invariant metric, an intrinsic definition of Cartesian stiffness is obtained by defining Hessian matrix of a potential energy function away from equilibrium. The resulting Cartesian stiffness is always symmetric even away from equilibrium.