This paper proposes a novel control method for stable grasping and attitude regulation of an object using a multi-fingered hand-arm system. This method is based on a task space sensory-feedback control using real time information of an object attitude. A unified control signal is designed for a whole hand-arm system. Firstly, nonholonomic rolling constraints between a multi-fingered hand-arm system and a grasped object are formulated. Secondly, a novel control signal to accomplish stable grasping and attitude regulation of the grasped object is proposed. To design the control signal, it is assumed that sensory information of the attitude of the grasped object is available in real time through some external sensors, such as vision, force, tactile sensors, and so on. Next, stability of the overall closed-loop dynamics and convergence of the object attitude to a desired attitude are given based on the passivity theory. Finally, several numerical simulations are conducted, and it is demonstrated that an arbitrary polyhedral object can be grasped, and its attitude converges to the desired one stably using proposed method.