Showing papers by "Keigo Watanabe published in 1993"

Mobile robot control using fuzzy-Gaussian neural networks

Abstract: A tracking problem for controlling the speed and azimuth of a mobile robot driven by two independent wheels is solved by using a fuzzy Gaussian neural network (FGNN) controller The computed torque control law is first derived to obtain some relationships between the driving torque and the plant output information To simplify the FGNN controller for the two-input two-output controlled system, a learning controller consisting of two FGNNs based on independent reasoning and a connection net with fixed weights is proposed The effectiveness of the method is illustrated by performing the simulation of a circular trajectory tracking control

Generalized pseudo-Bayes estimation and detection for abruptly changing systems

Abstract: The problem of state estimation and system-structure detection for linear discrete-time systems with unknown parameters which may switch among a finite set of values is considered. The switching parameters are modeled by a Markov chain with known transition probabilities. Since the optimal solutions require exponentially growing storage and computations with time, a new method of generalized pseudo-Bayes algorithm (GPBA) is proposed to circumvent this problem by using a multi-stage measurement update technique. A minor modification is also presented to correct a defect of the Jaffer and Gupta method. Some simulation comparisons are included to illustrate the effectiveness of the proposed algorithms. It is then shown that, as compared with other GPBAs, a feature of the present GPBA is that it noticeably decreases the size of the required memory when the number of states in the Markov chain is large. The cost to be paid is a slight increase in the computing time.

Fuzzy controller design using the mean-value-based functional reasoning

Abstract: A new fuzzy reasoning is proposed so that the conclusion consists of a function of mean-values on each membership function in antecedent. It is then shown that the constant parameters in conclusion can be rationally designed from the view point of a control theory, if the conclusion is regarded as a sliding controller with the mean values in antecedent as state variables.

Controls of servomotors for carry hospital robots

Abstract: Two methods for controlling servomotors with pulse encoders are presented for improving the movement of a carry hospital robot (CHR) along a desired line. Some simulation studies have been executed for the design of a PI controller or optimal regulator for the control of the DC servomotor. By applying the simulation results, we actually design a PI controller in an analogue circuit and experimentally compare the results of the PI control with those of a PLL (Phase Locked Loop) control. It is then clarified that the latter approach is effective in improving the accuracy of the CHR.

Nonlinear Control for Robot Manipulators with Artificial Rubber Muscles by using a Fuzzy Compensation

Abstract: This paper describes a nonlinear robust control for a robot manipulator with artificial rubber muscles by applying a fuzzy compensation. A fuzzy logic controller as a compensator is added to control the trajectory of a two-link robot manipulator, in which the computed torque control method has been already assumed to be applied to the robot. It is shown that when there exist model uncertainties and/or untuned feedback gains, the fuzzy compensator with a simple adaptive scaling technique is effective for the robust control. The effectiveness of the proposed control method is illustrated by making some simulations and experiments for the robot manipulator.

An Iterative Learning Fuzzy Control for Multi-Link Manipulators

Abstract: An iterative learning fuzzy controller is described for controlling the trajectory of a multi-link robot manipulator. The learning controller used here is called a multiple fuzzy controller, in which several elemental fuzzy controllers are processed in arallel and the degree of the usage of each inferred consequent is determined by using a linear neural network. Two learning algorithms are considered in the framework of the specialized learning architecture: one is based on minimizing the squared sum of the trajectory error for each link and the other is based on directly minimizing the squared trajectory error for each link. The effectiveness of the proposed control method is illustrated by making some simulations for a two-link manipulator.

Force Control of the Robot Manipulator with Collision Phenomena by Learning Control

Abstract: Impact/collision is a fast and non-linear phenomenon, so it is difficult to control a robotic manipulator undergoing collision phenomena. Therefore, in the past, manipulators were moved slowly in order to avoid collision. But with the recent increase in the amount of high-speed tasks, control of the manipulator undergoing collision has become indispensable. In such a situation, it is effective to use learning control in the forward manner. In this paper, we propose a new learning control method to optimize the weighted least-squares criterion of learning errors. This method can be applied to obtain a unique control gain, and it is shown here that the convergence of learning error can be readily assured. Using this learning control method, we carried out experiments on force control with collision phenomena, and proved the convergence of the output error. The robotic manipulator was made of an air-driven rubber actuator with no reduction gears to avoid damage due to impact.