In recent years, the researches for conducting the human motion capture and recognition are prosperously carried out in the field of man-machine interface, sign-language recognition, dancing behavior description and so force. Image processing techniques and magnetic sensing devices are widely used, but the former exhibits some disadvantages due to occlusion and the latter shows some problems in the spatial measurable range.
Therefore, the authors are proposing the measurement and processing techniques for dynamically estimating the human behaviors, especially the behaviors of rotary joint type link structures of upper arm motion by performing the integrated type sensor fusion using the gyroscopes and accelerometers having such special features that are free from the occlusion and are not influenced by the external magnetic fields environment.
This paper proposes a new measurement system which recognizes motion of human arms directly and precisely by the fusion of gyroscopic sensor and accelerometric sensor. That is, the objective of this system is the emphasis of the merits of both types of sensor and the removal of their demerits by unification of both sensors.
Experimental results using 1-joint and 1-axis arm show that the performance of this sensing system is acceptable within the range of human being's general behaviors, and the effectiveness of this method in application for human motion capture has been verified through the simulation using 2-joint and 2-axis human arm model.

View full abstract

Differential geometric approach is useful for solving a special class of nonlinear control problems. The purpose of this paper is to present a linearization method of nonlinear systems by the Riemannian geometric approach, and to apply the linearization mapping to the design of nonlinear systems.
A geomeric model can be derived by replacing the orthogonal straight coordinate frame on the state space with a suitable curvilinear frame. Such a Riemannian geometric model has been proposed by the authors after the derivation of the geodesic curve on the gravitational gauge field in Einstein's principle of general relativity.
In this paper an attention is placed on a problem to decrease the dimension of the Riemannian space of the model by a proper choice of the construction of the space, which leads to decrease the computation time remarkably. For the design of control systems, a new quadratic-form performance index is introduced using Riemannian metric tensors. A nonlinear optimal regulator is constructed which is homeomorphic to the corresponding linear optimal regulator. A method to derive a curvilinear coordinates frame fitted to the nonlinear system is proposed by solving a partial differential equation with respect to the homeomorphism. A computational algorithm is proposed and numerical examples are shown.

View full abstract

This paper presents control laws for multi-input nonholonomic chained system and two inputs three states nonholonomic Caplygin system. The control law for multi-input chained system is a smooth and static state feedback which exponentially stabilizes the equilibrium point except some special inital states. The Caplygin system is a class of nonholonomic systems and is widely applied in mechanical systems rather than the chained system. The two inputs three states Caplygin system can be stabilized by the proposed contol law which is piecewise analytic and time-variant state feedback.

View full abstract

In this paper, a new scheme to control nonlinear systems is proposed. A back-propagation neural network is applied to a nonlinear model reference adaptive control. Traditional model reference adaptive control techniques can only deal with linear systems or some special nonlinear systems. The back-propagation neural networks have the capability to learn arbitrary nonlinearity and are easy to apply to adaptive control applications. But, to construct a controller by a only neural network is very hard, because the charge of neural network is large and the learning speed of neural network become very slowly. Thus, for reducing the charge of neural network a scheme combining back-propagation neural networks with traditional model reference adaptive control techniques is proposed. Traditional model reference adaptive control is mainly employed to control the linearity of system, and the back-propagation neural network is employed as compensation of nonlinearity. Here, the employing traditional model reference adaptive control technique is dead-zone method. Moreover, in this paper the trasient response of the system become better by setting norm upper bound of a neural network compensator. Simulation results show that this scheme is validity.

View full abstract

The problem of innovations informational equivalence is discussed for observations with feedback. The additive noise is assumed to be a continuous square integrable martingale. Under the assumptions of 1) square integrability on the signal; 2) mutual independence between the noise and the part of the signal other than the feedback; 3) existence and uniqueness of the strong solution of the observation equation as a stochastic differential equation; 4) a weighted square integrability on the signal, it is shown that the innovations informational equivalence holds.

View full abstract

A method to construct a memoryless feedback law for systems with time-delay in states is proposed. A stabilizability condition by memoryless feedback is presented. A feedback law is constructed with a solution of a matrix equation. It is shown that the resulting closed loop system is asymptotically stable, and moreover, it is a linear quadratic regulator for some cost functional. The relationship between the stabilizability conditions presented in this paper and another is discussed. A numerical simulation is performed to evaluate the effectiveness of the control.

View full abstract

In this paper an LQR design method has been applied to develop a practical multivariable controller for an interactive flow and pressure control system of a thermal power plant. The low order controller which is obtained by the balanced realization has comparable burden of calculation and memory to conventinal controllers. The two degree of freedom configuration enhances the tracking ability. Experimental results in the utility power plant show the control performance.

View full abstract

In this paper, we deal with a control problem of a truck-trailer type mobile robot in backward maneuvering, where disturbances may act on its velocity. Firstly, we formulate the backward maneuvering problem as a functionspace minimax problem. Secondly, dividing the minimax problem into two subproblems-one is minimization problem and the other maximization one, we obtain its numerical solution. Lastly, we demonstrate by simulation that the control strategy obtained through the minimax approach is relatively robust to the velocity disturbances, compared with the strategy through the optimal control approach.

View full abstract

This paper describes an on-line load prediction system for supporting DHC (District Heating and Cooling) plants operation. Especially for DHC plants using electricity, it is desirable to shift electric power using from daytime to nighttime by thermal storage. For that purpose, it is necessary to make a schedule of the operation of DHC plant, based on the predicted hourly load for the next day.
This prediction system consists of three components. One is the daily total load prediction. Second is the load proportion prediction. The last is the hourly load predictions. Neural network model is applied for the load pattern prediction. This paper shows results of simulation on the load pattern prediction and the hourly load predictions using real DHC plant heat load data.

View full abstract

The products extruded from a granulating machine are easily smashed to powder by their physical characteristic, and their hydrous rate is changeable. So contact style quality analysis and batch analysis are difficult. Moreover, completely understanding quantitative relationship between many production factors and quality is never easy. On the other hand, production operator previously evaluated visually by knowledge based on his experience. So if a control system is considered, it seems to be effective to apply the computer vision methods on the quality estimation procedure. In this research, for the estimation, a texture is extracted from the product's surface, and each region of the product is recognized with extracting the contour using the steerable filter, simultaneously. And the classify to each quality is performed using a neuro-classifier, which learned from the teaching data of the operator's judgement on the quality.

View full abstract

In recent artificial intelligent technology, much attention has been focused on Behavior-Based AI. However, the arbitration between competence modules in this approach is still an open question. On the other hand, recent studies on immunology have clarified that the immune system plays important roles to maintain its own system against hostile dynamically changing environments through mutual interaction among lymphocytes or antibodies. In this paper, we attempt to construct a robot with a new decentralized consensus-making mechanism based on the biological immune system. To confirm the feasibility of our proposed method, we apply to a dynamic action selection for an autonomous mobile robot in simulated environment. Additionally, we try to evolve the proposed consensus-making network through genetic operations.

View full abstract

Evolution of an autonomous robot “Khepera” navigating in a course is studied, with a behavior model of automata and Genetic Algorithms. A set of infra-red ray sensor signals of the robot are inputed to an automaton which drives the two wheels. Any a priori maps or rules are not given to the robot. The transition table of the automaton is sliced into the chromosome and is evolved with Genetic Algorithms. It was demonstrated that a sensori-action model of this type can make the robot navigate successfully in a twisted course with obstacles and narrowness. The acquired behaviors, however, were found to be sensitive to the changes in starting conditions, where rugged fitness landscapes were observed. The fractal mechanism inherent to the wall-following navigation is proposed to explain the ruggedness. Averaging the ruggedness, the robustness was defined by the normalized average fitness. Several simulation experiments were carried out, where it was noticed that the robustness is lost after the major evolution is achieved. Several methods were tested, among which two methods; the fitness averaging over different starting conditions and the changing starting conditions during the evolution, were found to be effective in maintaining the robustness.

View full abstract

Backward movement control of articulated vehicles is one of the most difficult nonlinear control problems. Although the control problem of a mobile robot with one trailer has been reported in some papers, the control problem of a mobile robot with two trailers has not been completely solved yet. This paper describes backward movement control of a mobile robot with two trailers. Since the nonlinear model of the mobile robot can be represented by a Takagi-Sugeno fuzzy model, parallel distributed compensation is employed to design a fuzzy controller from the fuzzy model. Furthermore, we analyze the stability of the designed fuzzy control system via Lyapunov approach. The stability conditions for fuzzy systems are characterized in terms of linear matrix inequalities which result in convex optimization problems. It is shown that the stable fuzzy controller effectively achieves backward movement control of the mobile robot with two trailers in both simulations and experiments.

View full abstract

Deadbeat control achieves finite-time settling at sampling instants, but usually there remains error between sampling instants. In order to reduce the inter-sample error this paper introduces a quadratic performance index of continuous-time response and give a method for deadbeat control that minimizes the performance index.

View full abstract

This paper tries to improve estimations of angle velocity of servo systems. Angle velocity is usually carried out from angle data by observers, numerical differentiation, f/v transformation and so on. In this paper, we merge the all data derived by the above all methods, and we try to derive the best estimation. The experimental results show that this merging method remarkably improves the estimations comparing with ones by using each method separately.

View full abstract

In this paper we discuss a method which improves transmission characteristics of push-pull type shape memory alloy actuator. We propose the system which reduces driving force by the use of push-pull mechanism equipped with bypass. As a result, it is clear that this system is excellent about driving force and response than the old mechanism.

View full abstract