In this paper we consider the neuro-control method and its application to control problems of an electric ve- hicle. The neuro-control methods adopted here is based on proportional-plus-integral-plus-derivative (PID) control, which has been adopted to solve process control or intelligent control problems. In Japan about eighty four percent of the process industries have used the PID control. After deriving the self- tuning PID control scheme (neuro-PID) using the learning ability of the neural network, we will show the control results by using the speed and torque control of an electric vehicle.

抄録全体を表示

PID control is a common and an effective approach in process control represented by oil plants and chemical plants. However, it is important to determine suitable PID gains included in the PID controller because the control performance strongly depends on these PID gains. Lots of PID tuning methods have been proposed up to now, however, the computed PID gains cannot be employed directly for real plants. In addition, the system model is required in computing PID gains, and the accuracy of the model influences to the control performance. In this paper, a new design scheme of PID controllers without using the system model in computing PID gains. That is, PID gains can be derived by using only the operating (input/output) data. Concretely, the generalized output is newly derived from the PID control law, and the PID gains are computed by using the genetic algorithm so that the generalized output approaches to the system output. As a result, the suitable PID gains can be obtained, which the system output approaches to the reference model output. Finally, in order to illustrate the effectiveness of the proposed scheme, some simulation examples are discussed.

抄録全体を表示

PID control schemes have been widely used in most process control systems. As one of the reasons, the operator's knowhow can be easily utilized in designing controllers. However, most process systems are nonlinear, it is difficult to obtain good control performances using simple controllers with the fixed PID parameters. In other words, it is necessary to switch the most suitable PID parameters corresponding to system characteristics. Therefore, a PID controller design scheme is newly proposed for nonlinear systems. According to this scheme, the system model is first generated by using data-driven approach. Next, PID parameters are adjusted based on the internal model control(IMC) tuning. As a result, a suitable control performance can be obtained even for nonlinear systems. The behavior of the newly proposed control scheme is examined on a simulation example.

抄録全体を表示

There are growing interests for studying collective behavior including the dynamics of markets, the emergence of social norms and conventions, and daily life collective phenomena such as traffic congestions. In this paper, we deal with collective behavior in population of several social networks. We showed that collective behavior in cooperative relationship is affected in the structure of the social network and the initial collective behavior. We showed that collective behavior is affected by mediation coefficient but dimension of social network and diffusion of behavior is easy to emerge when mediation coefficient is 0.0. We showed that collective behavior is affected by clustering coefficient but topology of social network and diffusion of behavior is easy to emerge when clustering coefficient is less than 0.5.

抄録全体を表示

Multi agent simulation is useful tool for studying the interactions of complex systems. We choose combat simulation as an example of complex interactions. In this simulation, there are two types of interactions with environment. One is direct communication such as visible information and the other is indirect communication such as sharing the information with other agents. We investigate the effects of miss-perception during the communication and find that in some circumstances, miss-perception make the difference in the simulation results.

抄録全体を表示

A search method of the complex-valued spreading sequences that exhibits good correlation property for the code division multiple access (CDMA) communication, is investigated in terms of Genetic Algorithm (GA). The search problem of the spreading sequence is formalized as the combinatorial optimization problem that maximizes the fitness function for correlations. To solve this problem, a search method of suboptimal solutions based on simple GA is designed. As results, obtained suboptimal solutions are evaluated by their correlation characteristics and bit error rate (BER) for the CDMA communication.

抄録全体を表示

Usually, it is so diffcult to record conversation that it has been diffcult to conduct communication research quantitatively. Recently, people use Internet as media for communication.The conversation through the Internet is convenient for quantitative communication study because it can be recorded easier.Especially, Internet forum is suitable for collecting a various data of conversation. In this paper, lexical analysis is applied to the log data of conversation on Internet forum. It is very famous that many regularities in a document have been found, for example, Zipf ' s law, Heaps's law, and so on. This kind analysis has been applied to documents on a various of media. However, it is little studies that this analysis was applied to a document which is temporally developed by many authors, for example, the log data of conversation on Internet forum. Moreover, it has not been examined whether Heaps' law held good in the document and there is no discussion about the relation between the document size and these known regularities. Usually, the relation between document size and these regularities is not important because the size of such a document is determined by its author which is normally only 1 person. However, the size of the communication log of Internet forum is an emergent property by people who are interested in its forum. We believe it is important for us to understand the dynamics of conversation. As a result of investigation in this paper, the following 2 trends are found;(1)law of Zipf holds well in almost all log data of Internet Forum, (2) the number of messages posted is small if the vocabulary growth parameter ! of Heaps' law is not within a particular bounds.

抄録全体を表示

We have been developing a reinforcement learning technique called BRL as an approach to autonomous specialization, which is a new concept in cooperative multi-robot systems. BRL has a mechanism for autonomously segmenting the continuous state and action space. However, as in other machine learning approaches, overfitting is occasionally observed after successful learning. This paper proposes a technique to sophisticatedly utilize messy knowledge acquired using BRL. The proposed technique is expected to show better robustness against environmental changes. We investigate the proposed technique by conducting computer simulations of a cooperative carrying task.

抄録全体を表示

Diversification in finance is the process of spreading investments in heterogeneous asset classes. We provide a novel approach for evolving the diversification process by variable size Genetic Relation Algorithm(vs-GRA). Simulations using assets in USA, Europe and Asia indicate that the proposed approach offers competitive advantages for the global asset allocation problem.

抄録全体を表示

A new fuzzy rule mining method based on Generalized Genetic Network Programming(Generalized GNP) has been proposed to extract important time related association rules from sequential numerical database. The fuzzy set theory is applied to the data handling of continuous data in this paper. A new classification method based on extracted rules is also proposed to predict the future traffic density of each road in road networks. Experiments on traffic density prediction are carried out using the proposed methods and the results show that the proposed methods are available to find a variety of rules from the large database effectively and efficiently and improve the classification accuracy.

抄録全体を表示

A directed graph based evolutionary algorithm called Genetic Network Programming (GNP) has been proposed and extended with reinforcement learning algorithm (GNP-RL). Previously, GNP and GNP-RL had been applied successfully into the generation of programs for controlling mobile robot behavior and good training results were obtained. This paper aims to analyze the robustness of GNP-RL in the testing phase. For this purpose, programs for wall-following behavior of Khepera robot are generated by using both GNP and GNP-RL. In order to measure the robustness, variations in the operating environment is needed. In this paper, faulty sensors in the testing period emulate the change of the operating environment. The robustness of the evolved program is important to cope with the new situation. By using the reinforcement learning algorithm and ε-greedy policy in the testing period, a relatively good robustness against environment change is achieved by GNP-RL with proper parameter settings. After the environment changes, GNP-RL is still able to obtain rewards although the sensor(s) supplies wrong input values. GNP-RL has the flexibility to automatically and easily adapt to operating environment changes. It is done by selecting alternative actions which are prepared during the training period. In the simulation part, Khepera robot is simulated on Webots. The results obtained from the training and two different testing mechanisms are presented and analyzed. Additionally, the effects of different parameter settings of GNP- RL in the training and testing phase are analyzed to find the optimum settings in a given environment.

抄録全体を表示

Genetic Network Programming (GNP) is a graph based evolutionary algorithm. The efficiency of GNP, especially in dynamic environments, has been proved in many applications due to its directed graph structure. In this paper, GNP is applied to extract rules in stock trading markets. The method is an extension of GNP with Rule Accumulation (GNP RA). There are two main points: rule extraction and action determination. Rule extraction which is carried out in the training period is to extract the 1st order rules and 2nd order rules. Rules obtained from the individual with the highest profit are saved in different rule pools according to its action and order information. So far, in GNP RA, only the 1st order rules were extracted. Then, in this paper, actions are determined by the matching degree checking which order rules, the 1st order rules or 2nd order rules, are more useful. Simulation results show that the 2nd order rules perform better than the 1st order rules.

抄録全体を表示

With the increasing usage of Internet for business or personal affairs, the topic of internet security is becoming an extraordinary important issue. Therefore, many techniques for intrusion detection have been studied in order to build a secure internet atmosphere. In this paper, a new classification model, which is called Distance-based Classification Model, is proposed under the framework of GNP-based Class Association Rule Mining. The model uses the concept of calculating the distance between a data and rules in each class to classify the data. The simulation of evaluating this model is carried out based on KDD99Cup database.

抄録全体を表示

Generally, the mal factor for speech recognition is the background noise. As the noise is the main cause for decreasing the performance, the place or environment is very important in speech recognition. To improve the speech recognition performance in the real situations where various extraneous noises are abundant, a novel combination of FIR and Wiener filters is proposed and experimented. The FIR filter first selectively passes through the frequency range of human voice or speech, and then the Wiener filter filters out the extraneous noises. The combination resulted in improved accuracy and reduced processing time, enabling fast analysis and response in emergency situations.

抄録全体を表示

This paper proposes a new approach for designing LCL filter parameters for renewable sources by using intelligent optimization method. Generally, the renewable sources connected power converter system can cause high-order harmonics with the fundamental frequency around the switching frequency. On the other hand, LCL filter is known as an effective method on reducing harmonic distortion at the switching frequency. However, design of LCL filter requires effective ways to achieve good filtering effect because it is not simple to choose the proper LCL filter parameter. This paper presents an design method by using advanced bacterial foraging optimization and a new object function. In order to verify the performance of the proposed method, a photovoltaic system modeled by MATLAB/SimPowerSystems is used in this simulation. The proposed method showed more effective results than conventional methods.

抄録全体を表示

The conventional electro-mechanical circuit break and relay are widely used in large-sized DC power system. However, recently due to high reliability, remote controllability and small power dissipation of a RSSPC (Remote Solid State Power Controller), high-friendly DC power systems have increasingly adopted a RSSPC as a essential power switching device. In this paper, we propose a new RSSPC module which is implemented by using a DSP with parameters based on the simulation.

抄録全体を表示

If one can find a control Lyapunov function (CLF) for a given nonlinear system, the control input stabilizing the system can be easily obtained. To find a CLF, the time derivative of an energy function should be negative definite. This procedure frequently requires a control input which is a rational function or includes an inverse function. The control input is not defined on the specific state-space where the denominator of the rational function is equal to 0 or the inverse function does not exist. In this region with singularities, the trajectory of the control system can not be generated, which is one of the most important reasons why it is hard to make the origin of a nonlinear system be globally asymptotically stable. In this paper, we propose a smooth control law ensuring the globally asymptotic stability by means of cancelling the singularity in the control input.

抄録全体を表示

In this paper, we propose the best traffic control mechanism for network-based intrusion prevention system. We performed a test for it in Linux-based system. We considered various queueing algorithm as a traffic controller such as Class Based Queueing (CBQ), Stochastic Fairness Queueing (SFQ), Token Bucket Filter (TBF), Hierarchical Token Bucket (HTB) algorithm. Among the representative five kinds of queueing algorithms, TBF is the best performance in throughput and latency aspects. And we suggest CBQ and HTB are remarkable ones as the good traffic controller in our network-based intrusion prevention system.

抄録全体を表示

There have been various studies on ship collision avoidance. This research paper analyses a decision-making situation a navigator using the Non-Cooperative Game Theory.

抄録全体を表示

In this paper, we introduce the notion of fuzzy (r; s)- S1-pre-semicontinuous mappings on intuitionistic fuzzy topological spaces in Sostak's sense. The relationship between fuzzy (r; s)-pre-semicontinuous and fuzzy (r; s)-S1-pre-semicontinuous mapping is discussed. The characterizations for the fuzzy (r; s)- S1-pre-semicontinuous mappings are obtained.

抄録全体を表示

We consider fuzzy complex valued functions and fuzzy complex valued fuzzy measures, and Choquet integrals, and fuzzy complex valued Choquet integrals. The aim of this paper is to discuss some basic properties of a fuzzy complex valued Choquet integrals and to prove some convergence theorems for them.

抄録全体を表示

The recent concerns over the threat of global climate change and the requirements of national reduction of CO2 emission have led to the diversification of energy resources and a large scale integration of renewable resources. For island systems, however, the penetration level of renewable energy should be limited to maintain the power system security perspectives to frequency and voltage stability. The main interest of this paper is the operation of Jeju island system with a high penetration level of wind energy. As an option to make a breakthrough against the technical constraints, this paper presents a smart algorithm to set the active power of HVDC interties, which are to supply the power deficit and to regulate the frequency of the system.

抄録全体を表示

As the wind has become one of the fastest growing renewable energy sources, the key issue of wind energy conversion systems is how to efficiently operate the wind turbines in a wide range of wind speeds. Compared to fixed speed turbines, variable speed wind turbines yield higher energy efficiency, lower component stress and fewer grid connection power peaks. Generally, measurement of wind speed is required for the control of variable speed wind turbine system. However, wind speed measured by anemometers is not accurate owing to various reasons. In this work, an observer-based torque estimator and artificial neural network are utilized for estimation of effective wind speed. Finally, the performance of the proposed Maximum Power Point Tracking (MPPT) algorithm based on estimated wind speed is verified with simulation studies.

抄録全体を表示

This paper proposes a computational scheme for building a knowledge base of a real system, by measuring and processing the online data stream coming from the system. By assumption the real system under investigation is from the type of evolving systems that dynamically change (evolve) their parameters, structure and behavior with time. The proposed knowledge base is a kind of spatial-temporal model of the evolving system. It consists of memory cells that dynamically change by number (growing or pruning), contents and location in the input system space. Each memory cell makes weighted averaging of the output information, by using a memory buffer with a limited size that keeps the older information, arranged by a level of importance. The newer information is given bigger importance than the older one. All computational steps are discussed in the paper. A synthetic example of a 2-dimensional evolving system is explained in a graphical way (as gray-scale images) in the paper, for a better understanding of the whole computational technology. Finally some ideas for possible real applications of the proposed knowledge base are discussed.

抄録全体を表示

In order to visualize the electronic data of law, a twodimensional map model is proposed. The proposed system firstly performs morphological analysis of each law to generate input vectors, which are composed of extracted keywords from the laws. The proposed efficient Self-Organizing Maps method uses restricted region search and dimensionality reduction by utilizing the specific characteristic of the law data vectors, which have a lot of sparse elements. The obtained map represents the nature of relations of the laws relation nature which would be intuitively understandable to the users. To confirm the effectiveness of the proposed system, evaluation experiments are done by degree of similarity between nearest law using 150 law data from the year 2005 to 2009. The computation time of the proposed method is decreased into 64-98% that of the conventional learning method of SOM, and increase the precision of classification of target law data.

抄録全体を表示

Neural network as a data mining tool has a drawback in its black box nature, as the mechanism of relating the input and output parameter is not easily comprehensible by human. To extract the implicit knowledge from neural network in the form of if-then rules, the application of formal concept analysis is proposed. Using neural network trained with target data, a synthetic dataset of input and output is built and discretized to form a binary formal context. A formal concept analysis (In-Close) algorithm is applied to obtain formal concept, from where the implication rules are extracted. To show the effectiveness of the proposed method, an experiment is conducted using 480 data of six types of emotion and their corresponding HSV color components.

抄録全体を表示

We propose a new method to define the constants of Complex-Valued Q-learning, which is an Reinforcement Learning algorithm that can deal with incomplete perception problems. It enables agents without sufficient perception to recognize the context of actions at some level by applying complex numbers to value functions. We improve this method in two ways. One, we predict contexts not by adjacent situations as before but by how many times an agent acts from a starting situation. Second, we efficiently use memory that depends on the number of the steps that is required to get rewards. Our agents successfully solved more complex, incomplete perception problems by these methods. We also consider a context-based designing method for Complex-Valued Reinforcement Learning.

抄録全体を表示

The present study considers a fuzzy color system in which three membership functions are constructed on the color triangle. This system can process a fuzzy input (as the membership values of subjects) to an RGB system and output the center of gravity of three weights associated with respective grades. Three input fuzzy sets (red, green, and blue) are applied to the color triangle relationship. By treating three attributes of redness, greenness, and blueness on the color triangle, an average color value can be easily obtained as the center of gravity of the fuzzy output. The subjects are required to determine whether the impressive word (or the color word) for each color is suitable. The response of subjects can be obtained as fuzzy sets (as the membership values of subjects) on the color triangle. The responses for impressive words (stimulus, appetite, concentration, relaxation, calmness, rest, and mild) and the responses for color words (red, orange, yellow, green, blue, purple, and pink) show mountain-like shape. The height of fuzzy set indicates possibility of a target color or degree of cognition for a target color and the expanse of fuzzy set shows vagueness, the fuzzy sets for a impressive word (e.g. stimulus) and for a color word (e.g. red) are compared with calculating an intersection of those fuzzy sets. The fuzzy sets for impressive words show large vagueness and small possibility rather than those for color words. The differences between impressive fuzzy sets and color fuzzy sets are described, and the relationship between the centers of gravity of fuzzy inputs and inference outputs for fuzzy inputs are shown in the present paper.

抄録全体を表示

In this paper, we propose an Interactive Genetic Algorithm (IGA) with tournament evaluation applying win-loss result presumption that obtains evaluations of individuals from multiple people. In tournament evaluation, there is a problem that IGA users have to evaluate the same individual several times. Therefore it is thought that motivation for the solution evaluation of the users deteriorates. To overcome this difficulty, we propose that applying win-loss result presumption based on evaluation result records of the users to tournament evaluation by multiple people. The effectiveness of the proposed technique is verified using a simulation that employs multiple numerical evaluation agents instead of human evaluators. The simulation results shows to reduce the number of the play that evaluation agents voted for 89.0% at the maximum. Furthermore, we performed the performance comparison that assumed the conventional method comparison technique. The performance comparison results shows that an equal absolute grading value was provided even if the number of the play that evaluation agents voted is decrease.

抄録全体を表示

An internet service of video sharing using pseudo synchronization attracts the attention. As a typical example, there is NicoNico Video that comments which were submitted to any scene in a video by the users are superimposed to the same scene in the video played back by the audience. The audience watching the video with the comments feels as if they share the emotion with the users who submitted the comments even though they watch the video by themselves. In this paper, an influence of the superimposed comments to the audience is discussed through experiments in the emotional intensities of the audience and the usefulness of the video sharing using pseudo synchronization for the emotion sharing via the Internet is shown.

抄録全体を表示

The purpose of this study is to realize intuitive personal computer (PC) operation to reduce the gap between the real world and the virtual world by using a desktop environment with images of pictorial supplies and radio frequency identification (RFID). A preliminary experiment was conducted to obtain a suitable combination of supplies corresponding to actions on a PC. After analyzing the results, we developed a desktop environment where users can start application software by overlaying some pictorial supply images. We also developed a system using RFID that links a printed document to its file data in the PC. A subject experiment was performed to evaluate the proposed system. The results indicated that the system was effective for a user interface (UI) for novice users.

抄録全体を表示

A topic of learning autonomous robots has widely been discussed in these years. Reinforcement learning (RL) is one of the popular methods to this domain. Its performance is, however, quite sensitive to the segmentation of state and action spaces. To overcome this problem, we have been developing a new technique, Bayesian-discrimination-functionbased RL (BRL). Furthermore, BRL has proven to be more effective than other standard RL algorithms in dealing with multi-robot system (MRS) problems. However, as most learning systems do, BRL occasionally suffers from overfitting. This paper introduces an extended BRL for improving the robustness of MRS. Metalearning based on information entropy of firing rules is adopted for adaptively modifying its learning parameters. Computer simulations are conducted to verify the effectiveness of our proposed method.

抄録全体を表示

Evolutionary multiobjective optimization (EMO) algorithms have attracted much research interest in recent years. In evolutionary robotics (ER), several papers have been published where EMO algorithms have been applied to design multiobjective behavior of autonomous robots. However, these are for either specific control tasks or controllers. Characteristics of EMO algorithms on design of a more popular controller for simple robot control tasks need to be investigated for fully understanding them in ER. In this work, a multiobjective genetic algorithm was applied to the design of a neural controller for multiobjective behavior of a mobile robot in a looping maze problem, which is a popular test problem for ER. Distribution of non-dominated solutions in the objective function space were obtained from a number of trials in the problem in order to investigate how preferred solutions are distributed in them.

抄録全体を表示

Reinforcement learning approaches attract attention as a technique to learn mapping functions between sensorsmotors of an autonomous mobile robot through trial-and-error. However, traditional reinforcement learning algorithms need to prepare state space and action space that are appropriately divided by a designer beforehand. This paper proposes a new reinforcement learning algorithm that can learn mapping functions between continuous state space and grid action space, and the mapping functions are described by neural networks. The proposed method has two features. One is that the number of middle neurons and the initial value of weight parameters of neural networks are set through the dimensional number of state spaces automatically. The other is that an action is selected by using Griddy-Gibbs sampler. The proposed method is demonstrated through a navigation problem of an autonomous mobile robot, and comparing with when a designer sets the parameters of neural networks heuristically.

抄録全体を表示

One of the techniques for constructing an adaptive multi-agent system on the basis of evolutionary robotics (ER) is to generate robot behaviour suitable for an environment. However, even through careful examination, it is difficult to understand the strategies and features that are indicated by robot behaviours. Therefore, we have focussed on the environmental information used for the behaviour of robots. In this study, we attempt to construct an analytical method by using the recognition of a change in the information that affects a robot behavioural strategy. We propose a functional circle that describes the abstract functional relationships between robots and objects in their world for analysis of perception mechanisms and show the world of recognition by using self-organizing maps (SOM).

抄録全体を表示

A hybrid optimization based on evolution strategies and particle swarm optimization is applied to the car racing problem. In this hybrid algorithm, individuals of each population are divided into two groups by fitness. The first (and better) group is based on evolution strategies , and the second group is based on particle swarm optimization, which are chosen mainly to take advantage of the balance between exploring and exploit. In order to find out the performance of the hybrid, the computer experiment is tested on both a standard test function set and a car racing problem.

抄録全体を表示

In this paper, extended experience-repository-based Particle Swarm Optimization (EERPSO) is proposed for effectively applying Particle Swarm Optimization (PSO) to evolutionary robotics applications. It is an extension of our previous algorithm, called experience-repository-based Particle Swarm Optimization (ERPSO) which has the fast convergence property. The ERPSO uses a concept an experience repository to store previous position and fitness of particles of the PSO algorithms to accelerate the convergence speed. EERPSO additionally has an experience repository manager that selects particles to be included in the experience repository, and improved estimated best position selection mechanism in the ERPSO that balance exploitation and exploration of the EERPSO. We applied the EERPSO to find parameters of a gait of a quadruped robot for producing a fast gait. The EERPSO showed best performance among the original PSO, PSO and PSO variants, and ERPSO.

抄録全体を表示

In this paper, we propose a new approach for digital image watermarking in DCT domain. A texture analysis is applied on each 8x8 DCT blocks in order to classify the block as texture or non-texture region. The Ant Colony System Algorithms (ACSs) are then performed to find out the optimal parameters for watermark embedding, independently for each region type. The experimental results show that the proposed approach yields a watermark that is invisible to human eyes and robust to various image manipulations. The algorithm was also compared with the previous work, and our algorithm is superior for all test images.

抄録全体を表示

When a large-scaled disaster occurs, it is expected that plural lifelines get damaged at the same time. This causes big damage to our everyday lives, so we urgently need a plan to recover from the disaster. However, since lifeline networks have mutual relationships in terms of their functions, so it is difficult to promptly make a proper disaster restoration schedule. With this background, in this study, we aim to efficiently make a disaster restoration schedule with genetic algorithm taking account of the functional constraints among lifeline networks.

抄録全体を表示

Genetic Algorithms are the probabilistic search technique inspired by Darwin's theory of evolution. But there are many hypotheses in the theory of evolution, and there is the virus theory of evolution in one of them. Virus Evolutionary Genetic Algorithm (VE-GA) is the technique that is based on that theory. VE-GA is performed a global search and a local search by co-evolution of two kinds of populations. In late years Estimation of Distribution Algorithm (EDA) is studied actively. However, in general EDA, there is no mechanism that searches for the local solution. So it proposed an Estimation of Distribution Algorithm by partial solution in co-evolution. In the proposal method, partial solutions are generated by estimated value and search local solution by co-evolution using two kinds of populations. However, the consideration how it functioned was insufficient. We report that apply to several optimization problems and consider the factor that works effectively.

抄録全体を表示

This research was aimed to present the methodology of Ant Colony System for exam proctor schedules used in educational institutes. The purpose of the study is to cut down time for organizing each staff's proctoring table. Every staff has equal of proctoring during normal and extra workload. It is also able to avoid exam proctoring on weekend and on the staff engaged duration. In the experiment, it was found that the staff's proctoring table derived from method of the Ant Colony System can be applied to arrange the lecturers? exam proctoring tables effectively. The result is the proctoring tables that have the numbers of exam proctoring on the normal and extra workload are equal. It is also able to avoid exam proctoring on weekend and on the staff engaged duration.

抄録全体を表示

Actor-critic approach and natural-gradient-based methods have recently drawn significant interests in the area of reinforcement learning, and several algorithms have been studied along the line of the natural actor-critic strategy. This paper considers the problem of improving a previously reported RLS-based natural actor-critic algorithm toward a version that employs learning rate adaptation. In the actor part of the studied algorithm, we follow the strategy of performing parameter update via the use of the natural gradient together with learning rate adaptation, while in its update for the critic part, the recursive least-squares method is utilized for estimating the advantage function and the state value function. The applicability of the studied algorithm is illustrated via locomotion of a two-linked robot arm.

抄録全体を表示

In this paper, a development of magnetic position meter for unmanned ground vehicle or automated vehicle system using two-wired-interface(TWI) communication is presented. Usually, a magnetic position meter is installed under the bumper of an automated vehicle to detect magnetic markers or magnetic wire. The previous developed magnetic position meter is composed of several sub-boards with magnetic sensors, which use the parallel communication with each board. In order to overcome the limitation and minimize the size of the magnetic position meter, a serial communication should be applied. In this paper, we apply TWI communication to the board of magnetic sensors. The developed magnetic position meter is installed on the automated vehicle, and shows the high performance in the automatic driving test.

抄録全体を表示

A map generated from ground-based threedimensional LIDAR data is a critical component for autonomous vehicle navigation using a vision based sensor. When the size of a map is large and the number of grid cells is relatively big, managing the map associated with a dense data set from 3D LIDAR scanner is a demanding task. Wavelets serve as the basis for an efficient compression scheme which makes it possible to significantly reduce processing effort to generate and manage a grid map in real-time. This research proposes a novel approach to generate an occupancy map from compressed measurement signals. A one-dimensional Haar wavelet transform has been applied to compress 3D scan data, from which occupancy maps have been generated. Our experimental results show that this method performs well to provide an autonomous vehicle with rich 3D environment information. As an alternative method to reduce 3D scan data, Compressive Sensing is also adopted in this research. Compressive Sensing uses a small sub set of data to recover the unknown original data. Utilization of Compact Sensing might lead us to the development of a favorable sensing system; however, conventional compressive sensing recovery algorithms might require a non-trivial amount of computing. With sparse Bayesian learning algorithm, compressive sensing can provide a fast and cost effective solution for acquisition systems characterized by large data set.

抄録全体を表示

In the study, self-organization using color recognition is presented to overcome the required constraints in conventional self-organization using an external ceiling camera and wireless communication among robots like a robot soccer system. In this scheme, a singular association is introduced that a swarm robot considers only visual information from its reference robot or a moving target within LOS (Line Of Sight). In particular, this paper presents a set of cases that may occur in self-organization of swarm robots. The simulation results of prey pursuit show that the proposed self-organization can effectively construct self-organized swarm robots with the capability of migration even though a simple association rule by LOS.

抄録全体を表示

In this paper, we make a different authentication method according as processing technologies of various devices. And we suggest an adaptive device authentication framework that is performed single or group authentication by users needs. With this, we can provide safety of all devices and an efficient authentication method to users who use intelligent home network.

抄録全体を表示

An expert system is an important tool that a machine can substitute for an expert through the design of a knowledge base and inference engine using the experience or knowledge of an expert. Further, in the real world, a complex system requires synthetic estimation with the input of experts in various fields for the correct estimation of the situation, not any one expert. In particular, synthetic estimation is more important to navigation situations than in other cases, because of diverse potential threats. This paper presents a method of knowledge fusion pertaining to navigation safety knowledge from various expert systems, using a blackboard system

抄録全体を表示

In this study, methods to classify advertising reviews from shopping mall reviews are suggested. Advertising reviews are mostly written by companies and contain advertising contents. There are a few studies regarding the classification of opinion spam documents, which is very rare in foreign studies; however, there are no studies that classify advertising reviews from Korean reviews. In this study, the Naïve Bayes Classifier was used to classify review documents and the POS-Tagging and bigram methods were used to extract specific words. The frequency calculation methods for the probability value of specific words were: 1) the general number of appearances of words 2) the frequency calculation of specific words through the suggested Latent Semantic Analysis (LSA), and by recalculating the result from 1) in 2), the performances of each method were compared. As a result, the methods from 2) showed 88.4% accuracy which is 5.3% higher than 83.1% which was the previous result from using the POS-Tagging+Bigram method. Therefore, it was proved that the method suggested in this study is effective at classifying or extracting advertising reviews from Korean product review documents. I

抄録全体を表示

A legacy smart card which adapts a platform of Java Card loads and executes an application when electronics power is provided. However, recently the most smart cards are embedded into a mobile terminal as like USIM cards, therefore the power is able to supply for the smart cards. In this case, operation of a smart card using Java card system needs to consider its operating system to be enhanced in memory management, object management and transaction mechanism. In this paper, we suggest an advanced Java Card operating system which is able to have efficient installation, load and execution of application by applying a new memory management of the smart card that has safe power system.

抄録全体を表示

With the development of regional economy and logistics, ports within an area involve fierce competition. This paper discusses the application of game theory on port competition to help managers analyze the selection of strategies and response. Above all, through the economic analysis of port logistics market structures and supply-demand characteristic, we could see that the port industry shows a regional oligopoly situation. Therefore, game theory would be a helpful method on the analysis of competition and cooperation strategy between ports. And then we will present two game models conducted on port competition and simply talk about the equilibrium solutions. In our further research, more referential and well-established game models will be applied to the strategy analysis of port competition.

抄録全体を表示