This paper proposes a method for trouble detection from mechanical vibration.This method makes it possible to detect a parameter change in a dynamical system only from observed vibration signals without recourse to additional sensors.This is achieved by means of a newly developed blind system identification technique.

View full abstract

We develop the real-time speech visualization system called ``KanNon" which supports speech communication of hearing-impaired people. The KanNon system presents information of the speech such as characters including voice volume, pitch by speech recognition and sound spectrogram in real-time. We apply novel spectral estimation algorithm combinig Burg's method and minimum cross entropy method (Burg-MCEM) to the KanNon system. The Burg-MCEM estimates model parameters using prior information of AR model parameter based on the MCE principle. Further, we propose Burg-MCEM with distinction of prior information using distance of the Kullback information.

View full abstract

We are developing CMS that can do e-learning even in the scale at the elementary schools, junior high schools and high schools level, and this system can be introduced without efficient hardware and complex operation work. User's miscellaneous function is located and developed with an important function in development. In this research, the relation between the decided number of teaching materials and the level of the teaching material is modeled from a condition necessary for making the study plan. The parameter and the condition of the model describe whether a constant relation consists. As for developed CMS, it is scheduled to open it to the public as open source based on GPL with the teaching material made when developing when it reaches constant completion.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

In this paper, we propose a load-balancing algorithm for the parallel rule extraction system from trained neural networks. For efficient parallel computing in the rule extraction system, we need to reallocate fuzzy input vectors to each node in a PC cluster. Our proposed algorithm works very well to compute efficiently on several types of PC clusters.

View full abstract

A rescue robot contest seeds (ResCon-Seeds) propose the athletic meet which assumes the rescue by the robot to be a theme. This paper describes the outline, philosophy and possibilities of the rescue robot contest seeds. The chance to think about the society of the robot is offered setting the theme corresponding to the region, and holding small robot control. Accomplishing an educational contribution to a low age group in the region becomes possible.

View full abstract

When a great earthquake occurs in urban areas, it is important to gather information of a disaster areaas soon as possible.Especially, 3-dimensional terrain information of the disaster area will be useful to make a correct and optimal decisionfor rescue activities.Using an aerorobot with the laser range finder, the system which collects 3-dimensional terraininformation quickly and correctly is developed.Illustrative experimental results are shown in this paper.

View full abstract

This paper presents a method for cooperative control of a mobile robot between human operators and computers, where the system supports some cooperative action between manual and automatic modes. The fuzzy rules to avoid obstacles are adjusted by a turning algorithm based on human intervention operations. The method, proven by some actual trials in a PWS mobile robot, realizes efficiencies in the operation of a mobile robot with simplicities of commands and flexibility of human operator.

View full abstract

In use of flexible manipulators, interaction with environment such as collision or contact with an unlooked-for obstacle is inevitable. Damage avoidance of the manipulator is a significant problem in industry and space programs. This paper presents a position and collision control for a single-link flexible manipulator. Introducing a scalar (collision detection) function based on an innovation process, a collision detector is constructed. Then a twofold controller is constructed for rotating the manipulator to the desired position in normal operations and interrupting its motion when the collision is detected.

View full abstract

This paper proposes a new path teaching method for a mobile robot using a digital camera. In this method an operator takes pictures at essential points on robot's paths by a digital camera. Then, the operator teaches the robot the key points to recognize the images such as directions of the movement and some landmarks. By using this method, it is possible to teach a path for the mobile robot more easily. In this paper, a result of fundamental experiment at outdoor environment is described.

View full abstract

Autonomous Underwater Vehicle (AUV) is an underwater robot which executes the given instructions without operator's operation and judgment. Therefore, it is necessary to equip all the instruments including the source of electric power in the pressure vessel. So, generally AUV receives severe restriction in the action time and the action range. As solution of this problem, we chose the method that an AUV dock to an underwater station and charge its battery. And we developed the AUV "MARINE BIRD" which realizes it. This paper describes the feature and main functions of "MARINE BIRD", and the result of the sea trial .

View full abstract

A fear and oppression of the robot should be reduced in order to realize spatial and temporal coexistence of the human and the robot, for example, welfare and care robot, pet robot, personal robot, and so on.This paper describes about the applying a inflatable tube patched piezoelectric films to a robot arm component and the detection method of the contact position for the robot and the human.The objective of this paper is the evaluation of the relation between contact position and sensor output because of the sensing information for inflatable tube utilized the avoidance action of the robot.The experimentation were performed to confirm the variation of resonance frequency for the step external input at the position of the smart inflatable tube.The experimental results demonstrated that resonance frequency of the smart inflatable tube varies wit the contact position.

View full abstract

In this paper, we propose a multiple-SOM's (self organizing maps) approach for classification tasks. The proposed approach generates n versions of learning data set by processing an original learning data set. A map is trained using each of the versions, and hence n maps are obtained. Each map brings a result of classification. A global classification is made by means of majority decision of such results. This scheme is applied to defecting confusion between blood samples of different patients. A GA (genetic algorithm) is employed to processing original learning data. Experimental results show that the proposed scheme achieves higher accuracy of confusion defection, compared to the case of detection made by a single map.

View full abstract

Stanley and his colleagues proposed a neuroevolution method, called NEAT, that evolves both recurrent neural network (RNN) topologies and weights simultaneously. Experimental comparisons show that the method is much faster than previous ones on some challenging benchmark problems. The generation alternation model employed by NEAT, however, has a drawback; it proceeds without intensively searching neighborhood 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

Neutral networks, which are landscapes containing neighboring points of equal fitness, have attracted much research interest in recent years. In this paper, we applied the steady state GAs with different deletion methods to a simple model of neutral networks in order to investigate the effects of mutation rate on the speed of population movement. The result demonstrates that the speed of population movement has an optimal mutation rate and an error threshold since plotting speed against mutation rate results in a concave curve.

View full abstract

Neutral networks, which occur in fitness landscapes containingneighboring points of equal fitness, have attracted much researchinterest in recent years.In a recent paper, we have shown that, in the case of the simple testfunctions, the mutation rate of a genetic algorithm is an importantfactor for improving the speed at which a population moves along aneutral network. Our results also suggested that the benefits of the variable mutationrate strategy used by the operon-GA become increasingly clear as theruggedness of the landscapes increases. In this paper, we conducted a series of computer simulations with adifficult evolutionary robotics problem showing phases of neutralevolution in order to investigate whether our previous resultsare applicable to this problem. It is found that those evolutionary dynamics are consistent with our previous results. The variable mutation rate strategy is also beneficial with thisproblem.

View full abstract

[in Japanese]

View full abstract

This paper proposes a method to solve a dynamic layout problem with different areas of facilities using Evolution Strategies. Feasibility constraints are incorporated into the fitness function by introducing a penalty term corresponding to overlapped areas of any two facilities. The proposed method consists of two phases, such as a rough search phase to get an initial feasible layout and a precise search phase to obtain a final promising layout. Some numerical experiments made clear that the proposed method provides a solution with the same accuracy as the existing method, reducing computation time drastically.

View full abstract

[in Japanese]

View full abstract

We have proposed a generation alternation model for co-evolutionaryoptimization of strategies in one-to-one competitive games. In themodel, however, the populations of strategies are often trapped intolocal optima. To remedy the drawback, we propose an extended version of the model, which explicitly maintains diversity of the populations.The experimental results show that the proposed model performs betterover the previous one on a benchmark problem.

View full abstract

In evolutionary game theory, to the best of our knowledge, individuals' perceptions have been disregarded. When an individual interacts with the other individual under coexistence of heterogeneous populations, we can consider that the individual changes its strategy depending on the opponent. In such a situation, each individual may make an error about the population the opponent belongs to, and the influence of individuals' perceptions has to be taken into consideration.In this paper, we propose a multipopulation model with erroneous perceptions for opponents. We define an evolutionarily stable strategy (ESS) and formulate replicator dynamics in this model, and prove several properties of the proposed model.

View full abstract

To effectively optimize neural networks, we have proposed a co-evolutionary neuro evolution (NE) scheme, which selects an appropriate set oftraining examples adaptively for the current population of neuralnetworks, based on co-evolution of neural networks and examples. Thescheme performs better than the conventional NE ones, but it oftenloses the optimal networks identified previously, during the course ofthe co-evolution. To remedy the drawback, we propose an extendedversion of the scheme and show the effectiveness of the proposed scheme through its application to a function approximation problem.

View full abstract

In this paper, we examine structures of partnership that agents who maximize their payoffs form from the viewpoints of stability and efficiency. To do so, employing an artificial social model in which we can incorporate restrictions on partner selection, variations of the type number of agents, and the environmental changes, we perform the agent-based simulations for analyzing the partnership structures.

View full abstract

In this paper,an application of reinforcement learning to a radio-controlled helicopter is considered.An Actor-Critic algorithm is employed as reinforcement learning.A larening agent is composed of Actor and Criticthat are realized by using RBF neural networks.In the 3-dimensional model,a manipulation of helicopter needs 4 control signals for each of which one learning agent is assignd.Through computer simulation,it is observed that the agents learned flight controls to keep constant target altitude stably,however it couldn't learn flight controls of hovering.

View full abstract

In reinforcement learning problems, the learning of action-value functions is carried out in an incremental way. To realize a stable learning in such situations, we have proposed an RBF network model with memory mechanism. This model learns based on a gradient descent method. The gradient learning method is generally slow. To improve this problem, we apply the linear method to the learning of RBF networks. In this algorithm, the centers of basis functions are not trained, but only connection weights are trained. In our computer simulations, it is verified that the proposed model can learn fast and accurately.

View full abstract

Reinforcement learning(RL) attracts attention as a technique of realizing autonomous systems. However, in general, it is not easy to put RL into practical use. This difficulties include the problem of designing a reasonable state space of an agent contains two requirements in trade-off: to reduce the search space as much as possible in order to make a learning process be fast and to keep the characteristics of the search space as much as possible in order to seek better strategies. In this paper, it is considered to satisfy the above requirements. We newly introduced a concept of a "state space filtering". Then, we proposed an acquisition method which is to adjust the search space adaptively by referring to an entropy. The validity and the potential of the proposed method were comfirmed by computational experiments.

View full abstract

This paper proposes a synthesis method of an optimal supervisor in terms of a language measure by using a reinforcement learning.Recently, a concept of the language measure is introduced to the formal languages and a synthesis method of an optimal supervisor based on the language measure has been proposed.In this paper, we apply the reinforcement learning as a learning method of the language measure, and show that the optimal supervisor in terms of the language measure can be derived through learning.By computer simulation, we examine an opimality of the obtained supervisor.

View full abstract

An approach to controlling autonomous mobile robots that move around in flocks is described. Our main concern is how to design an on-line behavior acquisition mechanism which is capable of developing the cooperative behavior in an embedded environment. To design a behavior-learning controller, reinforcement learning is carefully implemented. The validation of our approach is done through computer simulations.

View full abstract

We have proposed a design scheme for asynchronous multi-agent systemsbased on neural netwowk representation of the decision policy and itsoptimization with a real-coded GA. To obtain global optima with thisscheme, however, substantial computational resources are required. Weneed a new scheme that allows us to obtain acceptable local optima with much less resources. In this paper, we consider (1+1)-ES as analternative of the real-coded GA and discuss its potentials andlimitations through the application to the seesaw-balancing problem.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

Research has seldom proposed concerning mathematical models of creative cognitive process such as insight or intuition of human brain activity. In this paper, the author proposes a new Lorenz model with two parameters, the temporal coefficient c and the spatial coefficient d. This spatiotemporal coupled Lorenz model with an excitatory coupling matrix or an inhibitory coupling matrix, which consists of c and d, is a device that synchronizes three 1-dimensional information codes. The c and d are parameters independent of each term's vector, and on-off intermittency observed in this model is controlled using the c and d. Results indicated that the presented model can be used as a device for description of emergent process.

View full abstract

A coupled system of phase oscillators is proposed for a signal control of an urban traffic network. An area-wide control is given by a network of autonomous systems, which exchange local information without any central monitoring system. Each signal is modeled by a simple phase oscillator with a periodic motion, and pair-wise interactions with the adjacent signals control its offset to attain a desired value. The interaction strength depends on a traffic flow density between the corresponding pair of adjacent signals. In this paper, the method is extended into a signal with multiple phases each for a traveling direction, especially for right turn traffic. The computer experiment by a traffic simulator is given using the extended method, and the average travel time is shown to be effectively decreased compared with the other offset values.

View full abstract

To study phenomena in which functions governing change of states should be regarded as variable, we introduce a shift dynamics called a functional shift which can describe dynamic change of functions.By using functional shifts, we describe dynamical behavior similar to chaotic itinerancy that is an itinerant motion among varieties of ordered states through chaotic motion.We analyze such dynamical behavior in terms of truncated Levy distribution, and ascertain the existence of itinerant dynamics both in variables and functions with a computer simulation.

View full abstract

This paper analyzes the operational skill for the search activity via a remote-controlled mobile robot to find out some hidden objects under debris. The experiments revealed that the skillful (i.e., both accurate and effective) operations require the adequate view control of the camera, which enables the parallel execution of the robotic ambulatory movement and the accurate environment recognition, ensuring the necessary perceptual resources. We also discuss some system properties required to realize the effective ``shared'' control with the machine-automation for human supports based upon another experiment after the analysis.

View full abstract

In this information society of today, it is often argued that it is necessary to create a new way of human-machine interaction. In this paper, an agent with social response capabilities has been developed to achieve this goal. In order to implement social intelligence in the agent, a mind model has been developed to render affective expressions and personality of the agent. The mind model controls the social behaviors of the agent, thus making it possible for the user to have more social interactions with the agent. The experiment with the social intelligent agent-based learning system suggested that a greater degree of not only learning performance but feelings of users was achieved.

View full abstract

We discuss the use of genetic programming for evolving social behavior of agents situated in inherently cooperative environment. We use predatous-prey pursuit problem to verify our hypothesesis that relatively complex social behavior may emerge from what we consider as Occam's razor for interactions: implicit, locally defined, and therefore-robust and scalable interactions between the prodator agents. The results show that surrounding behavior, evolved using proposed strongly typed genetic programming with exception handling emerges in successful team of agents.

View full abstract

[in Japanese]

View full abstract