Recently, embodied cognition for robotics has been discussed, and various types of artificial neural networks are applied for behavior learning of robots in unknown and dynamic environments. In this paper, we propose behavioral learning based on a fuzzy spiking neural network to realize high adaptability of a mobile robot. However, the behavioral leaning capability of the robot depends strongly on the network structure. Therefore, we apply a genetic algorithm to acquire the network structure suitable to the changing environment. Finally, we discuss the effectiveness of the proposed method through experimental results on behavioral learning for collision avoidance and target tracing in a dynamic environment.
Multiple Constant Multiplication (MCM) circuits are often used in Digital Signal Processing (DSP), such as Finite-Impulse Response (FIR) filters, linear transforms and so on, and a synthesis method to minimize computational complexity is required considering hardware cost. In synthesis of an MCM circuit, shift-chain made from partial sum is very important. The shift-chain greatly depends on the order of synthesis coefficient because it is updated in the process of synthesizing the coefficient. In this research, we propose a synthesis method considering the order of coefficients to be synthesized using the Genetic Algorithm. Furthermore, this method can reduce not only the number of adders but also the number of adder-steps. Through simulations, we show that the proposed method achieves better performance than the existing methods.
In this paper, we attempt reducing an error norm in structurally balanced truncation. Structurally balanced truncation is a controller reduction method developed from balanced truncation (a model reduction method) and is carried out by solving Linear Matrix Inequalities (LMIs). In this method, it is easy to get an a priori error bound which depends on LMI solutions, however, it is difficult to obtain optimal LMI solutions because the optimaization problem has the objective function which is not convex. To overcome the difficulty, a suboptimal procedure is proposed by Zhou et al., however, their procedure can not decrease the a priori error bound sufficiently. So we propose an error norm reduction algorithm utilizing a linearized objective function (a linearized equation of a non-convex function associated with the error bound) and balancing matrices. Finally, the validity of the proposed method are verified by using numerical examples.
This paper deals with a control method for a manipulator mounted on a free-floating space robot. A control method using the transpose of the Generalized Jacobian Matrix has been proposed. In contrast to many control methods using the inverse or pseudo-inverse matrix, the method guarantees the stability of the control system in the face of a singularity. This method, however, is developed on the assumption that all the parameters of the robot are known. That is, it is not sure whether the system stability can be ensured in the presence of parameter uncertainties. In this paper, we develop the adaptive control version. It is shown that the adaptive control system developed for the unknown parameters has the same stability property that the control system developed for the known parameters has.