The research for chaotic sounds is performed in the field of Artificial Life as one of the interactive art generation system. It becomes possible to create a music, which are generated by the computer, with the diversity and complexity exceeding the anticipation of human. The research purpose of the chaotic sounds is finding the effective method of making complex and variable sounds by using the chaotic theory. However, playing the corresponding sounds of a simple value generated by the chaos is expected to create the discord like a noise for human. Therefore, in this research, we propose a method of generating sounds by using the network of chaotic elements which both a chaotic asynchronous and a whole synchronism are controllable, and a method of incorporating some musical factors to be felt more natural music in the generated sounds. In this system, the musical note, duration and volume are decided by Global Coupled Map (GCM), and the additional music factors such as a track, a tonality, a rest, a tempo, an echo, and a timbre are applied. The created sounds are expected an effect of the healing music like a natural sound and an environmental music. We constructed the Interactive Chaotic Amusement System (ICAS) by applying this method and also report results of the Kansei evaluation carried out to confirm the effectiveness of a comfortable feeling for the sound made by this system.

View full abstract

We propose a method to aid a haptic device optimization task using Interactive Evolutionary Computation (IEC). The haptic device allows a user to touch 3D objects in virtual environments. The proposed method enables the user to intuitively create a touchable 3D object without demanding a specific knowledge about the device and any programming skills. The user touches a dozen 3D models generated by the system and rates them based on his/her subjective preference. Then, the system evolves the models using genetic algorithm (GA) based on the rates. A set of the user's subjective rating and the system's simulated evolution is iterated until the user gets a satisfactory result. We developed both an IEC-based haptic optimization tool and a manual configuration tool, and conducted a comparative experiment to validate the effectiveness of the proposed method.

View full abstract

In this paper, the authors propose a shape generation method based on the personal Kansei. The method works through an interaction between the user and a computer system. Several methods have been proposed to externalize “Tacit Kansei”, which is difficult to externalize the objective information such as words, although the user notices in his/her mind, up to now. On the other hand, “Latent Kansei”, which the user does even not notice in his mind, has not been discussed. The authors focus on Latent Kansei. The purpose of the method is to help the user to evoke the Latent Kansei by noting user's points of attention. A computer system generates new design samples based on the points of attention selected by the user. Reduct calculation in rough set theory is applied to estimate the features. The user externalizes both of Latent and Tacit Kansei through the iteration of the interaction process: generating design samples, showing the user's point of attention and evaluating design samples by the user. The system was compared to a conventional design system in which the user adjusts the design parameters by him/herself, and effectiveness of the system is demonstrated by experiments of shape design.

View full abstract

This paper introduces a method for case based learning of human assessment knowledge, and it is applied to a prediction of driving dangers. The human knowledge is represented by using decision trees to classify the situation described by attributes which are based on conditional information into an appropriate class. The practical application of driving dangers demonstrates usefulness of the tree learning algorithm GAD (Genetic Algorithm based Decision Tree Learning) for the human assessment problem in case of implementation. GAD is expected to involve a high generalization ability that it is robust to the lack and bias of the data. The driving situation experiment is consisted of cases which are represented by the attributes which can be defined as the location and the move of obstructions and the driver's judgment of danger. We apply the proposed method with GAD to the driving assessment example after the definition of the attributes and the object ordering algorithm which defines the importance of the obstruction. The proposed method is considered in terms of the number of sample required for learning, the learning accuracy and the size of the result tree. We also demonstrate the comparison between the human assessment and the extracted knowledge by our method in certain situations.

View full abstract

An autonomous agent has a ranged view using the absolute coordinate system, where it can receive accurate information in the range but noting out of the range. This is a considerably artificial situation. In this paper, we propose a staged view in distance and direction using the relative coordinate system, where an agent receives accurate information in the neighborhood but only rough symbolized distance and direction and rough distinction of other agents in short and middle-distance areas. It reflects a human's view that it can see easily an object in the neighborhood but more difficultly in the longer distance and easily in the center direction but more difficultly in the righter and lefter directions. We show by a numerical experiment for the pursuit problem, a multi-agent's benchmark problem, that the agent with the staged view learns effectively using Q-learning.

View full abstract

This paper presents a method for evaluating subjective visual perception. The subjectively equivalent size between a circle and a square is evaluated by this method. In this method, the data of subjective visual perception are weighted, because they are vague, and, if there is a contradiction among these data, the weights are decreased. Using these weighted data, an individual's perception is expressed by a membership function. The typical perception of many people is obtained as a fuzzy set by aggregating these individuals' membership functions. This method does not require strong assumptions for the data and is able to show a lot of information of the subjective visual perception, comparing with statistical approach. Human interfaces, nowadays, are demanded to provide subjective satisfaction for users and hence this method is effective and practical.

View full abstract

In this paper, we propose a route choice behavior model of the car drivers in road traffic, which is introduced into our developing microscopic road traffic simulator MITRAM for investigating the effects of road traffic information on its traffic circumstances. The developed model has its struncture of connecting some fuzzy inferences in the form of network, which is shown to be effective in modeling the route choice behaviors including some uncertain factors.
The model is verified using a benchmark test which is populor one. In two simulations by MITRAM, we investigate the effects of the differences in driver's decisions for route choices, and of providing and non-providing the drivers with the road traffic information on the traffic congestion. We show more wide applicability of MITRAM through the simulations.

View full abstract

Fuzzy c-Means (FCM) clustering is an unsupervised classification method for revealing intrinsic structure of multivariate data sets. It is, however, applicable to databases including only numerical variables. For analyzing the intrinsic feature of categorical data sets, the quantification of nominal variables estimates low dimensional scores. This paper proposes a new approach to the clustering of mixed databases including not only numerical variables but also categorical variables. The clustering technique uses an FCM-type simple iterative algorithm including a quantification step, in which the category scores are derived so that they are suited to the FCM clustering by considering cluster centers and memberships.

View full abstract

There are probabilistic restrictions on traditional Fuzzy c-Means methods which identify membership functions with the sum of membership values at each element as one. On the other hand possibilistic clustering methods identify membership functions without such a constraint, but the shapes of membership functions are independent of the clusters estimated through the possibilistic methods. In this paper, we proposed FCM with regularization by confusion degree. Using our method, we can obtain non-additive membership functions, and their shapes depend on the data distribution, which means that they differ from each other. To show the feasibility of the proposed method we have carried out some numerical experiments.

View full abstract

For the efficient management of call centers, fluctuation of the number of calls should be considered in planning operator's schedule. A scheduling designed as to satisfy sufficient service levels at the busy hours, cause over-staffing at the vacant hours. The operator scheduling problem for call centers is complex, and has huge search space, has considerable difficulties in making good scheduling considering the variable amount of calls. In this paper, we applied GA, parameter-free GA, TS and the combination of PfGA and TS to this problem, and compared their effectiveness. It has turned out that these proposed methods enables us to make an optimal scheduling considering the trade-off between costs and service-levels using a quite simple evaluation function. A series of computer experiments shows effectiveness of the proposed methods.

View full abstract

In this paper, we propose a new framework which can be applied to a major class of reinforcement learning methods. It enables autonomous robots to obtain behavioral concepts incrementally through on-line interactions with their environments and rewards. This framework is based on J. Piaget's schema theory, that puts emphasis on the Co-existence of the two processes ; assimilation and accommodation, and equilibration and differentiation. This approach is aiming at the realization of a social robot which can obtain many behaviors through interactions with its users and its environments. Our framework can be applied to any TD-learning methods. This paper presents the results of two experiments. The first one deals with Q-learning, and the other one deals with Dual-Schemata model based reinforcement learning. In both cases, agents obtain some behavioral concepts without any explicit indications about differences between those behaviors by their supervisors. Moreover, it is shown that rewards to learning robots are given a new role as recalling the most suitable schema.

View full abstract