[in Japanese]

View full abstract

In order to visualize the position of decision units with multi-inputs and multi-outputs, this paper proposes to combine data envelop analysis evaluating the performance of them and correspondence analysis positioning them in two-dimension graph. Adding another dimension for performance to the graph by using Mathematica, we have verified the availability of the proposed technique for the sample data.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

In GAs, the fitness evaluation is one of the most important part.There have been reported several studies about GA in dynamic or noisy environments which focused on the fitness evaluation.However, few researches have been reported which pay attention to the concept of fitness evaluation in general GAs.In this study, we discuss the concept of fitness evaluation and propose the speedup technique of GA by means of improved fitness evaluation.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

Genetic Algorithms (GAs) are the effective optimization method.There have been proposed lots of GAs, and each GA is specified by operators and neighborhood structure. However, it is very difficult to recombine the different GA's operators or intoroduce a new neighborhood structure. To solve this problem, in this study, we propose a flexible GA framework.We also propose the GUI support tool for constract a new GA in propsed framework.

View full abstract

This article presents a multi-agent search algorithm for function optimization, conducted by multiple agents. Those agents perform local search tasks repeatedly, updating their starting points according to the previous results, and attempt to identify optimal solutions collectively. During the course of the collective optimization, some agents merge into a single one to suppress redundant search processes.

View full abstract

[in Japanese]

View full abstract

Recently, the number of day trader has increased rapidly. On the other hand, many researches of forecast of stock prices have been based on the closing price,there are few researches which focus on the day trade.In this paper, the authors proposed the system of getting stock priceand discuss the strategy evolution of day trade agent by means ofthe evolutionary computation. We also propose a novel support toolfor day traders.

View full abstract

This paper presents a composite machine learning architecture of imitation learning and reinforcement learning. Reinforcement learning usually requires many trials and errors in an agent's learning phase. However, people can reduce the amount of time by imitating other people's way of performing the task. Therefore the composition of reinforcement learning and imitation learning is proposed as an integrated machine learning architecture.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

When a function of leg feet declined in aging disorder and convalescence, we need walk training. In this case we are convenient if we get up from a reclining bed and can walk-training with posture as it. Therefore we suggest the walk-training robot which can do walk-training by competitive walk on a bed side. We explain a walk-training robot that can detect a movement course from landmarks by neuro network.

View full abstract

A closed loop system identification method using measurable disturbances without adding probing signals to the control input has been developed. The proposed method identifies the deterministic subsystem and stochastic subsystem by using the orthogonal decomposition and a subspace method, and derives a plant model by utilizing algebraic relations between the two subsystems.

View full abstract

In this paper, the authors propose a method of tuning of the user-defined parameters for the distribution-based 4SID algorithm developed for continuous-time stochastic systems. Proposed method decides the parameters for a good estimate by using the correlation analysis between the input-output data and the test function.

View full abstract

We present an identification algorithm for continuous-time Multiple-Input Multiple-Output systems from the samples of the input-output data. In the algorithm, from the sampled data first an equivalent discrete-time model is identified, next the model is transformed to the corresponding continuous-time model. Then the choice of the sampling interval plays a key role.

View full abstract

Evolutionary Computations(ECs) are the optimization methods based on the mechanism of natural evolution. Although several studied have been reported on analysis of ECs, there are few researches which study an operator based analysis. In this paper, the authors proposed the novel analysis methodwhich based on operators. Computer simulation is carried out to show the advantage of proposed method.

View full abstract

[in Japanese]

View full abstract

Aimed at the establishment of Genetic Programming(GP)-based emergent systems design schemes, we have proposed a new framework of GP that can resolve one of the critical problems in GP i.e. the problem of explosive increase of complex individuals in the population. In this article, we discuss the potentials of the proposed framework through its application to the two-dimensional hockey game.

View full abstract

[in Japanese]

View full abstract

We have proposed a generation alternation model for co-evolutionary optimization of strategies in one-on-one competitive games. However, the model often degrades the best strategies in the population during the course of co-evolution. To remedy the drawback, we propose an improved model, which takes advantage of a long-term memory of past significant strategies much more effectively than the previous model.

View full abstract

The generation alternation model employed by one of the best neuroevolution methods, NEAT, has a drawback; it proceeds without intensively searching neighbors of parent individuals and hence it often fails in identifying solutions with minimal structures. We propose an alternative for the model and show its effectiveness through the application to the double pole balancing without velocities problem.

View full abstract

[in Japanese]

View full abstract

In this paper, we propose a ubiquitous Q-learning engine realized by a small-sized microcomputer. In order to reduce the convergence time of Q-learning, we introduce a user's instruction learning mechanism into the engine. As a practical application of the engine, we have developed a fan control device that is fixed on user's clothes and makes a confortable environment for user.

View full abstract

Cooperation of each participating agent is necessary in many multi-agent tasks: e.g., playing football, carrying a large table, etc. In cooperative tasks, an agent has to estimate the other agent's intention by watching how state variables change. In this paper, we describe a Situation-Sensitive Reinforcement Learning (SSRL), which enables an agent to notice the other agent's change in his/her intention and learn adequate behaviors.

View full abstract

[in Japanese]

View full abstract

This paper proposes a learning method for acquiring a dribble skill of a soccer agent. The learning agent mimics the behavior of a well-trained target agent. For this purpose, we utilize log files of soccer games where the target agent played to train neural networks. Through computational experiments we show that the learning agent obtain the dribble skill of the target agent.

View full abstract

Finite-dimensional modeling of heat conduction systems with frequency error bounds is presented where the finite number of eigenstructures of the systems are only partially known.Use of frequency response data is discussed in characterizingfeasible set of systems.

View full abstract

We consider a problem of desingning a closed loop system for an unknown plant based on input/output measurements. In this paper, we propose a direct method to obtain parameters of the desired controller via controller identification from the input/output measurements and the desired response instead of model dentification of the plant.

View full abstract

In this paper, under the assumption that a plant is linear and time invariant, we provide a sysnthesis of linear controller unfalsified the experimental data and achieves the desired specification.Moreover, we mention the relationship between VRFT, FRIT and the methods we proposed here.

View full abstract

In this paper, we propose a subspace identification method for multivariable systems including multiple input and output delays under the influence of colored noise. This method estimates an impulse response matrix from input and output sequences based on an orthogonaldecomposition-based approach. The simulation result shows that the delays and system matrices can be identified accurately by this method.

View full abstract

This paper considers a procedure of identifying Hammerstein models with white noise input, assuming that only a finite string of output data is available.A linear dynamical system is first identified by using the output data and then a static nonlinear function is estimated from the probability density function .A numerical simulation is also included.

View full abstract

Stochastic subspace identification methods have several integer parameters such as the model order and the number of block rows of data matrices. In order to give a guideline for choosing such parameters, this paper proposes a criterion for evaluating the models estimated by means of subspace identification algorithms. The proposed method is based on the power spectral density which is approximatelly computed from the given time-series data.

View full abstract

PCM synthesis is a method for synthesizing sound of musical instruments.By this method, a real sound of an instrument is recorded into digital data. The sound by PCM will be similar to the real one,however it needs very large memory size.We therefore introduce subspace identification for sound synthesis,and identify the sound generator system by using sound with noise.We show that our method needs less memory than PCM.

View full abstract

In this paper, we will present a gradient-type recursive subspace modelidentification algorithm which is applicable to identification of slowlytime varying systems. Using the proposed recursive algorithm, a changedetection method will be provided to detect changes in systems undersurveillance.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

To assess the sedation level of patients undergoing general anesthesia, many sedation indices based on the electroencephalogram (EEG) have been proposed. In this paper, we calculate basic elements for determining sedation indices using wavelet transform, and compare the results with those using short-term Fourier transform. Also, we propose novel basic elements using wavelet transform and study their availability for the sedation index.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

The function and performance evaluation of this teaching materials are considered to be necessity understanding of a rescue robot from the monodukuri literacy education viewpoint in this paper. This paper was mentioned the concept of the module type robot kit aiming at robot technology education and was considered the effectiveness.

View full abstract

[in Japanese]

View full abstract

This paper addresses the autonomous moving control of th symbiosis rescue robot with crawler arm in collapsed house. The algorithm of autonomous moving control with getting over obstacle was proposed and was performed some numerical simulations. Results of numerical stimulation was indicated that optimal main body attitude angle existed to every height of the approaching obstacle.

View full abstract

This paper is concerned with a modification of Real Time Kinematic Global Positioning System, called as a server type RTK-GPS.It enables us to accurately observe the movements of electric power poles in real-time at a distant surveillant station.The experimental results show that this positioning system can measure the three dimensional movements of poles with centimeter order's accuracy.

View full abstract