Journal of Japan Society for Fuzzy Theory and Intelligent Informatics
Online ISSN : 1881-7203
Print ISSN : 1347-7986
ISSN-L : 1347-7986
Volume 21 , Issue 5
Showing 1-36 articles out of 36 articles from the selected issue
Special Issue: Human Symbiotic Systems
Original Papers
  • Hiroshi SHIBATA, Masayoshi KANOH, Shohei KATO, Hidenori ITOH
    2009 Volume 21 Issue 5 Pages 630-639
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    To improve face-to-face interactions between people and the robot,“Ifbot”, we propose a facial expression control method that searches the path with the lowest facial changes in the emotional space created by an auto-associative neural network. Unlike methods based on linear changes of facial control parameters, our method can create delightful and humorous face changes. We report that this method shows characteristics that combine the linear changes of facial control parameters with the linear changes of coordinate points in emotional space. These results suggest that the facial expressions Ifbot creates using this method can increase diversity of interaction.
    Download PDF (1332K)
  • Yasutake TAKAHASHI, Yoshihiro TAMURA, Minoru ASADA
    2009 Volume 21 Issue 5 Pages 640-652
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Both self-learning architecture and explicit/implicit teaching from other agents are necessary not only for learning beahvior for a task but more seriously for life-time behavior learning. This paper presents a method for a robot to understand unfamiliar behavior shown by others through the interaction between behavior acquisition and recognition of observed behavior, where the state value has an important role not simply for behavior acquisition (reinforcement learning) but also for behavior recognition (observation). That is, the state value updates can be accelerated by observation without real trial and error while the learned values enrich the recognition system since it is based on estimation of the state value of the observed behavior. The validity of the proposed method is shown by applying it to a dynamic environment where two robots play soccer.
    Download PDF (1530K)
  • Hidekazu SUZUKI, Hitoshi NISHI, Koji TAKI
    2009 Volume 21 Issue 5 Pages 653-662
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    In the field of pet robots and robot-assisted therapy (RAT), characterization of animal motion is important for the development of robots resembling various animals. This paper presents a method for the generation of animal gait in quadrupedal robots. In this study, we employed AIBO as an experimental quadrupedal robot and generated the gait of the robot on the basis of an animal's gait. First, we optimized the mono-leg orbit, which can efficiently output a propulsive force, by imitating a dog's gait using a genetic algorithm. Moreover, we generated the quadrupedal gait of AIBO using both the optimum orbit of the mono-leg and an animal's gait, classified as the gait of a walking dog based on zoology. Furthermore, we administered a questionnaire study to determine subjective human feelings to choose the best gait for AIBO from among the various gaits mentioned above. Finally, minor deviation of parameters for each joint was corrected to realize the stable gait on the ground.
    Download PDF (1224K)
  • Shinya NAKAMURA, Naoto IWAHASHI, Takayuki NAGAI
    2009 Volume 21 Issue 5 Pages 663-682
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    In this paper, we propose an utterance generation method that will enable a robot to form a shared belief efficiently and in a mutually adaptive way with the user. The shared belief is formed based on a common experience between the robot and user. It enables inference of the state of each other's belief systems, as well as comprehension of some of each other's ambiguous utterances. In the proposed method, a belief system that a robot has consists of two parts: a shared belief function, which expresses the shared belief assumed by the robot, and a global confidence function, which represents the degree of coincidence between the user's shared belief and the robot's. The shared belief function is composed of a set of weighted belief modules each of that represents a concept such as motion, object, and spoken language and so on. The global confidence function outputs the predictive probability that each other's utterances have been understood correctly. The belief system is learned incrementally, online, through human-robot interaction with objects. By learning the global confidence function, the robot becomes capable of inferring the state of the user's belief system and develops a prediction probability, adaptively generating utterances according to the situation as well as to the degree of coincidence of the shared belief, e.g., the increase and decrease in the number of words. Through interactions with generated utterances, the robot can itself update the global confidence function; in this way, the user and the robot adaptively form mutually-shared beliefs. The validity of the proposed method is shown through a number of experiments under various conditions.
    Download PDF (1695K)
  • Naoyuki KUBOTA, Aiko YAGUCHI
    2009 Volume 21 Issue 5 Pages 683-692
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    The expectation to home partner robots has been increasing for the care of aged people and children. However, the communication system of partner robots is based on scenarios designed beforehand, which makes natural communication with people difficult. In this paper, we aim at realizing natural communication between people and robots by the learning of the relationship between perceived information and utterance contents. First, we propose a method based on spiking neural networks for the learning of the relationship. Next, we conducted several experiments on the learning of the relationship between perceived information and time-dependent utterance contents. The experimental results show that the robot can update utterance contents based on the cognitive development according to the interaction with the person.
    Download PDF (734K)
  • Kentaro ISHII, Yasuhiko HATO, Thomas KANOLD, Michita IMAI
    2009 Volume 21 Issue 5 Pages 693-700
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    This paper proposes a function introduction interaction, which encourages human-robot interaction on first contact with a self-driven interactive robot. The function introduction interaction aims to cause a person's interest by asking the person to watch the robot's interactive function. We developed a promotion robot and conducted a field experiment to verify effectiveness of the function introduction interaction, and analyzed interaction logs, and questionnaire answers. The results showed that the function introduction interaction enabled the robot to perform its task and decreased the feeling of being interrupted.
    Download PDF (493K)
  • Ryota FUJIMURA, Guo BIN, Ren OHMURA, Kazuhiro NAKADAI, Michita IMAI
    2009 Volume 21 Issue 5 Pages 701-712
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    In this paper, we describe a movable projection avatar system named “Remy”. Remy aims to support communication that share other party's actual environment from a remote environment. There are three issues in a lot of existing studies in remote communication system. First, there are few systems that consider sharing other party's actual environment. Second, even in the systems sharing other party's actual environment, users suffer from distraction of some devices. Third, nonverbal communication is not considered in a lot of existing systems. Remy solves these three issues by projecting two dimension avatar on actual environment. Conducting some Remy's evaluations from local user's perspective, the results showed that Remy can solve three issues and enhance quality of communication.
    Download PDF (1158K)
  • Kazuki KOBAYASHI, Katsunori KADOWAKI, Yasuhiko KITAMURA
    2009 Volume 21 Issue 5 Pages 713-721
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Life-like agents have the potential to make e-shopping sites on the Web more attractive and persuasive; our interest is to determine how multiple life-like agents should behave as a team to persuade customers. To know how the effectiveness of persuasion depends on the number of agents, we develop a multi-agent persuasion system. In the experiment, we compared the performance of persuasion by one agent, two agents, and three agents respectively. The experimental result shows that two agents agreeing each other deliver the best performance in persuasion. In addition, the result of the detailed analysis suggests that expressing the same attitudes among the agents toward the user delivers a better performance in persuasion than expressing the different attitudes among them.
    Download PDF (751K)
  • Yoichiro MAEDA, Satoshi HANAKA
    2009 Volume 21 Issue 5 Pages 722-733
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Generally, it is known that the engineering application simulated from the learning mechanism of animals is useful to make learn behaviors of the autonomous agents or mobile robots efficiently. Above all, a general idea of “shaping” used by ethology, behavior analysis or animal training is a remarkable method recently. “Shaping” is a general idea that the learner is given a reinforcement signal step by step gradually and inductively forward the behavior from easy tasks to complicated tasks. In this research, we propose a shaping reinforcement learning method took in a general idea of “shaping” to the reinforcement learning that can acquire a desired behavior by the repeated search autonomously. Three different shaping reinforcement learning methods used Q-Learning, Profit Sharing, and Actor-Critic to check the efficiency of the shaping were proposed and the experiment by the simulator of grid search was performed. Furthermore, we proposed the Differential Reinforcement-type Shaping Q-Learning (DR-SQL) applied a general idea of “differential reinforcement” to reinforce a special behavior step by step such as real animal training, and confirmed the effectiveness of this method by the simulation experiment.
    Download PDF (1509K)
  • Hidefumi OHMURA, Daiske KATAGAMI, Katsumi NITTA, Takayuki NOZAWA, Tosh ...
    2009 Volume 21 Issue 5 Pages 734-746
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Recently, researches of Human-Agent Interaction (HAI) are popular toward autonomous agents to act and cope in various human societies. If agents adapt to humans' society, they should have social skills. These can't be dealt with through conventional methodology of HAI. In this paper, we introduce how developing a social adaptive agent. There are many rules in societies. Therefore, social agents should have a social skill which is obtaining social rules. We develop agents adapting communities as social adaptive agents. We improve some functions from a simulation game: Online BARNGA from BARNGA to observe humans' behavior. We observe humans' behavior in Online BARNGA. As a result, we find that humans have transitions of three inner states with “notice” and “behavior” to get implicit rules, and rules fall into two categories: a byelaw and an ethic. In Online BARNGA, a byelaw is a rule of card game, an ethic is a strategy. We analyzed the brain activities of humans obtaining implicit rules by the functional near infrared spectroscopy technique. As a result, humans' blood flow was decreased at complained by others, and increased at thinking, and it was related to learning. Based on the above results, we develop agents with transitions of three inner states with “notice” and “behavior” and two type learaning modules. Furthermore, we analyze agents' behavior in Online BARNGA, and compare humans' that. As a result, we confirm that the agent adapts socially as if it were human.
    Download PDF (1389K)
  • Jun-ichi KUSHIDA, Iori NAKAOKA, Kazuhisa OBA, Katsuari KAMEI
    2009 Volume 21 Issue 5 Pages 747-756
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    The adaptive resonance theory (ART) neural network is an unsupervised learning system that can generate and grow the recognition categories based on the similarity between inputs and memories. Recently, several researches about Reinforcement Learning System which used ART as a construction of state-space have reported. Inthesystem, a proper vigilance parameter that decides a category space size is required for the adaptive learning in finite time period. However, the proper value of vigilance parameter is unclear on generally unknown in each problem. Therefore, this paper proposes a new learning system using hierarchical Fuzzy ART for two player games. The proposed system segments an input state space into subspaces by the fuzzy ARTs added hierarchically in proportion as the learning progress of player, and then learns pairs of input states and actions by the reinforcement learning. As results of experiments, we show through a fighting simulation game that the player could acquire the proper pairs of the input states and actions by learning using hierarchical Fuzzy ART against the opponent player.
    Download PDF (926K)
  • Hiroyuki INOUE, Dan YUAN, Kae IWATANI
    2009 Volume 21 Issue 5 Pages 757-767
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Various colors are used for things that we use in daily life. A color combination is important for design and has a great influence on image or human kansei. The aim of this study is to construct a color combination support system considering personal preference and human kansei. In this paper, we propose a IECs based color combination support system that shows color combinations that the user likes more. First, we construct a IECs based color combination evaluation system. In this system, the user can makes a favorite color in the beginning of the system first. This color is one color composed of three color combination, and the other two colors combined with this color are generated by IEC. Next, we develop the color combination evaluation system, and construct the IECs based color combination support system. Also, the sort function are introduced in this system in order to support user's relative estimation. Using these systems, color combinations of user's favorite was obtained. Also, effectiveness of functions of changing the position of the favorite color and sorting candidates were shown.
    Download PDF (1079K)
  • Yoichiro MAEDA, Shigeru MIYASHITA
    2009 Volume 21 Issue 5 Pages 768-781
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    The research purpose of interactive sound is to achieve the sound generation with complexity and variety which exceeds the forecast of human. By using the chaotic theory, it is expected that a new sound which affects human's sensitivity is generated. In our laboratory, the sound generation system has been developed using Globally Coupled Map (GCM) which many chaotic elements are put in order and the state is made to transit the whole interaction by passing through the averaging process. GCM, both chaotic asynchronism and whole synchronism are controllable, is able to generate various sounds. By adding some music elements in this system, we are able to generate the sound which does not give displeasure to human. However, there is a problem which payloads to human also become large in case parameters controlled by human operator increase. Therefore, in this research, we introduce the interactive genetic algorithm (IGA) which makes evolution guide to the direction corresponding to human's Kansei and aim for constructing a method to make sound generation easy. IGA is an optimization methodology united the human evaluation and optimization ability of genetic algorithm. ICAS is able to control the complexity of output sounds by only two parameters of synchronism and asynchronism, therefore, it was easy to apply IGA to ICAS. We constructed the simulator ICAS-IGA1 which adjusts only parameters of GCM automatically and ICAS-IGA2 including also parameters of musical elements, and conducted the experiment of sound generation. As a result, we confirmed that the sound generated by ICAS-IGA1 and ICAS-IGA2 coincide with human's Kansei in general. Furthermore, the comparison experiment of performance and operativeness between the conventional method (ICAS) and the proposed method (ICAS-IGA1) was executed to confirm the efficiency of this system.
    Download PDF (1101K)
  • Shunsuke AKIGUCHI
    2009 Volume 21 Issue 5 Pages 782-791
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    It is still in the way to make music generation system which reflects human sensitivity in various research organizations. Among systems which are studied, there are typical systems such as the system which composite music from the impression of words and the system which utilize Genetic Algorithm. The purposes of this study are to compose several music, to cope with the sensitivity of each individual by visualizing the relationship between human sensitivity and musical structure, quantify the whole musical structure which decides the impression of music and to realize the construction of the automatic musical creation system which reflects individual sensitivity by the method of modeling the relationship of this structure and each individual sensitivity by Neural Network. First of all, a whole musical structure to decide its impression shall be quantified as musical structure element such as key, harmony, melody, and pitch as well as rhythm and tempo based on the idea, “Music is composed of three elements such as rhythm, melody, and harmony”. Then, the relationship between music and impression felt from the music shall be replaced by the followings: the relationship between musical structure element quantified music and impression words quantified impression by SD method. Therefore, the relationship between music and its impression from music shall be learned. Secondly, the relationship between music acquired through learning and its impression shall be used to acquire musical structure element of impression levels to enter by Genetic Algorithm. This regards musical structure element as the individual of genetic algorithm. Therefore, this regards musical structure element as entries of learned Neural Network to output the value of musical impression of input music. To learn the difference between the value of musical impression obtained here and the value of musical impression to compose as assessment function automatically generates music of its impression.
    Download PDF (747K)
  • Yusuke KAJIHARA, Yoichiro MAEDA
    2009 Volume 21 Issue 5 Pages 792-803
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    In this research we propose a music composition system based on Twelve-tone Technique that is a music composition technique of the contemporary music, as the composition method by the computer. Twelve-tone Technique has the advantage that it is suitable for the composition of computer music because it is a technique to compose the music step by step. Making of the twelve-tone row, which is the first process of the Twelve-tone Technique, is an important process to determine the subject and atmosphere of the music. In this research, we design the fitness function based on the relation between the consonance interval and dissonant interval that are a part of general musical theory, and perform the automatic generation of twelve-tone row by searching the consonance tone row with GA. We constructed the simulator that generates twelve-tone rows and confirmed the effectiveness of this system by executing the comparison questionnaire of twelve-tone rows made by the simulator and human.
    Download PDF (1415K)
  • Haruhiko SHIRAI, Jousuke KUROIWA, Tomohiro ODAKA, Hisakazu OGURA
    2009 Volume 21 Issue 5 Pages 804-814
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    In this paper, we reported the results of experiments applying the user authentication method with fuzzy measure evaluation we had proposed to the Schonlau's data set. In the method an authentication system checks the current user's behaviors monitoring the command chains inputted by the user under interactive computer environment. We applied the method to the Schonlau's data set and analyzed the detection capability comparing with other methods. We evaluated the performance of the method with a ROC (Receiver Operating Characteristic) curve. According to the results of several experiments, the method reveals a good ability to detect intrusion at the level as some as the other methods such as the HMM method showed. Also, we discussed about the some characteristics of Schonlau's data set which seem to affect the detecting performance.
    Download PDF (547K)
  • Yoshiyuki MIYAGI, Naruaki TOMA, Satoshi ENDO
    2009 Volume 21 Issue 5 Pages 815-826
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Ontolopedia, a Japanese-language ontology dictionary system intended for general use, was designed and constructed using the text of the Japanese-language version of the online encyclopedia Wikipedia as the corpus. In the present study, we evaluated the application of a knowledge map created from this dictionary system for accurately extracting and predicting the interests of individual users. The interests of each user were extracted and predicted from comments acquired via Twitter using the knowledge map constructed from Ontolopedia. Only nouns were extracted from the acquired data. Using the proposed technique and prepared vocabulary lists, words of interest were extracted and predicted for the following three items. First, each word in the vocabulary list was assigned a numerical score based on level of interest, and higher rank for words that generate more interest by users was verified. In order to verify these points, we compared our method with the widely used TF-IDF method. Second, we verified whether extraction of interest words was possible or not. The point of this discussion was carried out a situation in which the interest words don't appear directly in the vocabulary list. Third, we considered whether user interest characteristics could be determined by investigating interest biases with classified concepts on the knowledge map. As a result, the three kinds of evaluations showed that the proposed method aorded greater accuracy in extracting words of interest compared to conventional methods. Furthermore, we showed that the user interest biases could be observed. This observation result will improve the accuracy in extracting words of interest.
    Download PDF (797K)
  • Takahiro TANAKA, Kinya FUJITA
    2009 Volume 21 Issue 5 Pages 827-836
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    In this paper, we experimentally discuss the duration of uninterruptibility reduction effect at focused Application-switching (AS) using Near Infrared Spectroscopy (NIRS) and subjective evaluation scores during PC operation. The NIRS results suggested that AS corresponds to the halfway of the concentration on the task using the switched application, and it takes five seconds on average to reach the concentrated state. Moreover, the subjective evaluation scores revealed that the interrupts within two seconds from AS show significantly lower uninterruptibility than the interrupts after four seconds or more. These results suggested that the AS uninterruptibility reduction is an acute effect that lasts in several seconds.
    Download PDF (600K)
Regular
Original Papers
  • Hisashi HAYASHI, Seiji TOKURA, Fumio OZAKI, Tetsuo HASEGAWA
    2009 Volume 21 Issue 5 Pages 856-869
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Adaptation to the dynamically changing world has been an important subject of research. Typically, two types of agents have been researched: deliberative agents and reactive agents. Deliberative agents “think” before and while taking actions. On the other hand, reactive agents react to the changing world without thinking too much about the future. Some researchers have been trying to combine deliberation and reaction because even deliberative agents need to react tentatively and immediately to the changing world when unexpected events occur. These kinds of agents are supposed to react within a short time without changing the current situation too much. However, in the case of an emergency, even while executing a plan for a goal, agents might need to achieve a more important goal for which “emergency deliberation” is necessary. If the agents suspend the current plan and execute the emergency plan, they might change the situation greatly and it might not be easy to continue the suspended plan. In this paper, we will combine deliberation and “emergency deliberation,” which is beyond the combination of deliberation and reaction. Because our agents continuously modify the suspended plans while executing an emergency plan, they can resume the suspended plans correctly and efficiently even if the world has changed greatly due to the emergency plan execution.
    Download PDF (844K)
  • Syuji KANEKO, Hakaru TAMUKOH, Kazuhiro TOKUNAGA, Tetsuo FURUKAWA
    2009 Volume 21 Issue 5 Pages 870-883
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    This paper aims to develop hardware implementation of higher-rank of self-organizing map (SOMn). SOMn provides a novel algorithm for multi-class learning scheme, which is expected to be a good foundation of intelligence of autonomous robots and mobile devices. For this purpose, it is important to implement SOMn on small energy consuming and small size VLSIs. In this paper, we have implemented SOM2 on a Filed Programmable Gate Array (FPGA). The most critical point was the number of logic elements, which restricted the size of the architecture, whereas other aspects such as memory capacity and calculation speed satisfied the required criteria. It was also shown that the logic element size problem could be avoided by using a latest FPGA device or modifying the circuit design. With considering these results, our SOM2 hardware on FPGA enlarges the potential of SOMn as a foundation of intelligent agents and devices.
    Download PDF (1329K)
  • Keisuke ICHIDA, Keigo WATANABE, Kiyotaka IZUMI
    2009 Volume 21 Issue 5 Pages 884-893
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    This paper discusses the control method of an underactuated manipulator, which has a property of nonholonomic systems. Since there are few actuators than the numbers of joints, the underactuated manipulator has various advantages, such as lightweighting, energy saving and cost reduction. However the underactuated manipulator needs complex control. Therefore we devised a fuzzy energy region method based on a logic based switching method. The fuzzy energy region method is a technique using the logical switching rule based on the energy region expressed with the fuzzy rule. We consider the projection configuration in the case of three links with two actuators, and discuss the property of these systems. Results of the present method are illustrated with some simulations.
    Download PDF (626K)
  • Qiang LI, Yoichiro MAEDA
    2009 Volume 21 Issue 5 Pages 894-904
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    Genetic Algorithm (GA) has been successfully applied in wide scope, and is a learning algorithm to mimic the biological mechanism of inheritance (neo-Darwinism). In general, because GA is an exploration method including stochastic search, there were a number of issues. Specially, the search ability of ordinary GA is not always optimal in the early and final stage of search, because of fixed genetic parameters, i.e., crossover rate, mutation rate and so on. Therefore, we have already proposed the fuzzy adaptive search method for parallel genetic algorithm based on the acceleration of evolution and high quality solutions. However, there are some cases when it is not enough accuracy to describe the stage of evolution, because the best fitness and average fitness were adopted as inputs of fuzzy rules. Moreover, worse performance was shown in the test function with high dimensions. Therefore, in this research we propose the improvement methods that have a good performance in the optimization problem of high-dimensional function. And the comparison simulations are executed to verify the efficiency of proposed methods. The results of simulations are also reported.
    Download PDF (1156K)
Short Notes
  • Osamu MORIKAWA
    2009 Volume 21 Issue 5 Pages 905-912
    Published: October 15, 2009
    Released: January 12, 2010
    JOURNALS FREE ACCESS
    D. G. Schwartz introduced the formal system Σ of subjective epistemic reasoning in which belief was measured along a series of linguistic degrees, e.g. unequivocally believes, strongly believes, fairly confidently believes, somewhat believes, neither believes nor disbelieves, somewhat disbelieves, fairly confidently disbelieves, strongly disbelieves, unequivocally disbelieves. The purpose of this paper is to present two axiomatizations for D.G. Schwartz's system, by using the matrix of many-valued logic. Then we prove the completeness theorem as well as the soundness theorem.
    Download PDF (525K)
feedback
Top