This paper demonstrates that a novel neural structure can be useful for formation of long-term memory. An essential core of the proposed model is a control signal for learning or memory formation, inspired by a possible biological learning mechanism observed in cerebral cortices and hippocampus. Simulation results show that the model can possess some biologically observed features of human memory systems.
This paper proposes neural network models for learning and identifying deterministic finite state automata (FSA). The proposed models are a class of high-order recurrent neural networks. The models are capable of representing FSA with the network size being smaller than the existing models. We also discuss an identification method of FSA by training the proposed models of neural networks.
A control system, which consists of several cooperative modules whose combination and structures change dynamically according to the state and environment, is discussed in this paper. We propose a method to design a control system by modular learning. Numerical simulations and flight experiment of an autonomous aero-robot demonstrate the effectiveness of the proposed method.
This paper applies a homotopy continuation method to improving the BP training of quasi-ARX neural network model. The idea is to start the BP training with the criterion function for linear ARX model, which is gradually deformed into the actual one for quasi-ARX neural network model. The effort is made to build the deformation into a usual BP recursive procedure.
In this research, we propose an application of the random-like feature of deterministic chaos for the exploration generator in reinforcement learning. In a no stationary shortcut maze problem, the chaotic exploration generator based on the logistic map gives better performances than the stochastic random one. In order to understand the differences, we investigate the learning structures obtained from the two explorations.
The descending two major pathways from a basal ganglia-brainstem-midbrain control the activity of muscle tone and the locomotion executing system. We have modeled a bipedal walking system with these two pathways and demonstrated that the bipedal locomotion is flexibly organized by integrating the two pathways, which is more adaptable to the changes of the environment than the other CPG models.
We have proposed Attentive Workbench (AWB), a new cell production system in which an intelligent system supports human workers. Using cameras, projectors, self-moving trays driven by 2D linear motors and so on, the system recognizes the worker's condition and intention and supports the workers from both physical and informational aspects. In this report, a method for controlling multiple self-moving trays keeping their formation is presented.
We have formulated and examined an autonomous organization system for transportation network system. We have considered costs not only to traverse but also to maintain some routes on a transportation network. The results say that the proposed algorithm changes a transportation algorithm into a feasible topology that has the smallest cost to traverse and the property of the small-world network.
The authors have proposed a dynamic turning control system of a quadruped robot by using nonlinear oscillators. It is composed of a spontaneous locomotion controller and voluntary motion controller. In this article, capability of dynamic turning motion of the proposed control system is verified through numerical simulations and hardware experiments.
In this paper, we consider a torque pattern learning during biped walking based on the feedback balance control. By learning, the torque profile of feedback controller is copied to feedforward controller consisting of the oscillator. Then, the ground reaction forces, which are essential to behave in uncertain environmental conditions, are not required. We simulate this process under some environmental conditions.
We propose a face detection method using compound features. Four kinds of features are combined to construct a compound feature vector. The projection of the feature vector on a reduced feature subspace learned by PCA is used as the input of the classifier, which is a polynomial neural network (PNN). The experimental results demonstrate the effectiveness of the proposed method.
In this paper, we propose a new 3-D face recognition system using 3-D face shapes obtained from stereo pair face images. The 3-D shapes are rendered under various illumination conditions, head-pose, scale and occlusion. The effectiveness of our system is demonstrated by the experimental results.
In our previous papers at the SICE2003, we proposed the complex-valued eigenface for the efficient representation of 3-D normal vector maps. But the uncertainty of pose/position of the face can decrease their performance. In this paper, we propose an automatic adjustment for pose/position, by placing one axis for symmetry and by choosing several control points based on Gaussian curvature.
The authors have proposed a super-resolution algorithm for polarimetric radar data that we call polarimetric bandwidth extrapolation (PBWE). In this paper, PBWE is further extended to two dimensional case. Two dimensional PBWE (2D-PBWE) utilizes two dimensional polarimetric linear prediction model. The performance of the 2D-PBWE is shown through a simulated and a real polarimetric SAR (Synthetic Aperture Radar) image.
This paper presents a novel method for moving object extraction using multi-tiered Pulse-Coupled Neural Network (PCNN). The multi-tiered PCNN model is called E-PCNN, since excitatory term and external linking are its two features. It is shown that E-PCNN outweighs the commonly used inter-frame difference algorithm, having three main advantages: utilizing multiple color information, parameter robustness and against noise.
Position and force simultaneous trajectory tracking controller with a pneumatic cylinder driving apparatus is proposed. The controller applied to the driving system is composed of a non-interaction controller and a disturbance. The experimental results show that the interacting effects of two cylinders are eliminated remarkably and the proposed control system tracks the given position and force trajectories accurately.
This paper presents a theoretical model of cylinder shell bulk modulus to achieve more accurate modeling of effective bulk modulus of oil in automotive hydraulic dampers. The theoretical model has been derived based on the elastic theory for thick-walled cylinder incorporating not only radial but longitudinal deformation. The new models on shell bulk modulus were examined through numerical computations.
The advantages of pneumatic systems have led to the development of Pneumatic Artificial Muscle (PAM) Manipulators. However, some limitations still exist, such as deterioration of the performance of transient response due to the change the external inertia load. To overcome this problem, switching algorithm of control parameter using LVQNN is newly proposed. The effectiveness of the proposed control algorithms is demonstrated through experiments with different external inertia loads.
The Yaoyorozu Project has been formed in order to design ubiquitous information society in 2010. This design problem calls for integration of humanity and engineering sciences. In order to foster collaboration among researchers in diversified disciplines, development of scenarios is emphasized. The present paper describes background and research architectures as well as summary of the research accomplishments so far.
The ubiquitous information society will be come by the development of technology in the future. To make the society safety and reliable, it is important to prescribe privacy guidelines and examine legal and ethical problems from an initial stage of technological development. This paper presents activity and study of POL (Policy and Ethics)-ST in the Yaoyorozu Project.
This paper presents significance of a systematic approach in the ubiquitous information society. The systematic approach is inevitable to manage the society, since it has too many decision makers of humans and machines. The approach consists of three steps as visualization, modeling, and optimization. It is shown how to apply the approach on validation of software as an example.
An evaluation of the effectiveness of a night driver support system providing advance notice of pedestrian location is presented. Experimental results indicate an increase of the driver’s visual attention on salient regions and a reduction of the number of collision by a factor of three. This demonstrates the benefit of the system in improving nighttime driving safety.
This paper is concerned with an application study of model-based fault detection method to a ship propulsion system. When modeling the nonlinear system, Quasi-ARMAX model with multi-model form is used and Kullback discrimination Information (KDI) is introduced as fault detection index. Some schemes to improve the fault detection performance are proposed together with simulation results for various fault modes.
Today's automobiles with petrol engines are equipped with 3-way catalytic converters that purify automobile exhaust emissions. In this paper. we construct a simplified model for the dynamics of 3-way catalyst based on the assumption that catalyst behavior is dominated by the dynamics of O2 storage. Then we propose a method for parameter identification of the constructed model.
The dielectric characteristics of agricultural land or soil are investigated by LCR meter. Four soil models consisting of clay, sand, organic matter, and their mixture were analyzed by sweeping the frequency from 100Hz to 2MHz. Each soil model has its own characteristic.
This paper describes recipe utilization and generation to analyze safety and support the operator in handling plant operation using embedded-batch recipe. Operators need supporting system to explore parameter, safety aspects and other important information during batch operation. According to batch standard, batch recipe can be performed and as references for ongoing processes. Process analysis can be carried out using Sensor Markup Language that is reasoned by recipe processor
An alcohol oxidase immobilized gas-sensor (bio-sniffer) with stick-type configuration was developed for convenient analysis of ethanol vapor. The calibration range of the ethanol sniffer covered the concentration range encountered in breath after alcohol consumption including the permissible legal limit for driving. As the physiological application, the bio-sniffer was used to monitor the concentration change of breath ethanol after drinking.
The paper examines some of the concerns in the design of systems composed of very large numbers of networked 'intelligent sensors'. The problems of configuring and maintaining such networks are examined, along with the system design issues raised. The cogent sensor is an essential part of realization of the type of applications envisaged. A systems architecture is outlined.
In this study, a minimum number of sensors will be applied on the side of two cylinders on the diesel engine in order to build an abnormal diagnosis system which detects anomaly behavior at an early stage by analyzing the acceleration waveform using the ALM method. This measurement system can be applied to the ship.
We have developed an eddy current based technique to achieve an online precise measurement of the surface silicon content of content graded 6.5% silicon steel sheets. The technique utilizes minute direct magnetization for permeability control of the surface to suppress the permeability change due to variations in the surface silicon content.
Measuring the cable tension force is important to cable-stayed bridges. According to frequency method, this paper advanced a new wireless measurement system based on LabVIEW. During signal processing, DTFT method was used to obtain precise fundamental frequency. Experiment has proved that the system can fulfill the fieldwork requirement of cable force measurement very well.
The authors have recently proposed a new signal processing technique called M-transform. In this paper, the authors propose a new method for impulsive noise reduction by using of M-transform and wavelet shrinkage.
An effective method is proposed for the estimation of instantaneous frequency of signals which are observed corrupted by random noise. The method is based on the idea of maximum likelihood function constructed indirectly from the Wigner distribution of the observation data. The effectiveness of the method is confirmed by simulation experiments.
This report introduces a new construction of sequences having both a low peak factor and a flat power spectrum. Since the proposed sequence has a flat power spectrum, its auto-correlation is zero except for the zero-shift. The proposed construction uses a systematic scheme and no search method. The length of the proposed sequence is n(4n-1) for an arbitrary integer.
This paper proposes a new design method of an I-PD controller in multirate system. A multirate I-PD controller is designed based on a multirate generalized predictive control law. The control effect of the multirate I-PD controller is greater than that of single-rate one. To show effectiveness of the proposed method, a simulation result is illustrated.
We propose a model predictive tracking control for wiener models using dynamic output feedback and adopt the Norquay's Wiener model predictive control algorithms for static output nonlinearity of the wiener model. We developed wiener model for air separation plant using subspace identification and applied to proposed control scheme.
This paper proposes a design scheme of two degree-of-freedom GPC for m-input m-output systems. The proposed method reveals the effect of the integral compensation only if there exist modeling error and/or disturbance. Therefore, performance degradation due to an integral compensation, such as a slow response or an excessive control effort, can be avoided when the controlled system has no perturbation.
Un-autonomous mobile robot has potential for a lot of practical applications. The main contribution of this paper is the development of new fuzzy command smoothing concept for smooth tele-operation of un-autonomous mobile robot with many obstacles around. It yields smoothed command for tele-operated mobile robot from the movement of force feedback joystick. HILS system developed in this research show the validity of the proposed concept.
A new fuzzy-logic-based self-tuning PI controller is proposed for speed control of indirect field-oriented induction motor. Here, a well designed fuzzy-logic provides the suitable gain of PI controller to eliminate the steady-sate error and overshoot which are occurred in conventional PI controller. The effectiveness of the proposed fuzzy-logic based self-tuning PI controller of induction motor was demonstrated through simulations.
This paper presents a model reference fuzzy adaptive controller approach in the design of fuzzy controllers to regulate fuel and air for the combustion process of utility boiler. The on line adaptive fuzzy model for fuel is presented by calculating the adaptive error factor considering steam pressure to the turbine inlet and turbine first stage pressure and for air is by measuring oxygen in the flue gas path. An experimental setup is fabricated and simulation studies were carried out.
Positive output Luo self-lift converter is a new DC-DC boost converter. Power electronic systems are generally complex and fuzzy control works well for these systems. This paper discusses the design and MATLAB simulation of a Fuzzy Logic Controller (FLC) for the above converter. The proposed FLC maintains the output voltage constant irrespective of line and load disturbances
In atmospheric wireless optical communication, the ambient may change stochastically, it may impair the communication, so the channel processing must be self-adaptive to guarantee high quality of the link, immediate processing is also critical for high rate receiving-transmitting and relaying. The novel Layers Active Orthogonal Phase Position Matrix technologies (LAOPPM) obtain good results, and it is simple to implement LAOPPM.
Design methods of adaptive control of uncertain nonholonomic systems are presented based on the notion of inverse optimality. The proposed methodologies are applied to chained systems of high orders, and the controller designs by both state and output feedbacks are shown. It is seen that the resulting control strategies are optimal to certain meaningful cost functionals.
This paper presents a hybrid adaptive control that adapts the closed-loop system to match with the reference model. The adaptive analog controller consists of a set of analog controller and a switching controller. The switching controller selects a controller from the set of analog controller and connects it into the closed loop controlled system suitably.
Speed regulator based on feedback of armature voltage and current was applied to squid jigging machines to avoid speed overshot. The key point is to identify the system parameters. Model reference adaptive system based on RLS identification was adopted. The design scheme was successfully applied to squid jigging machines and avoided the regulator adjustment respectively for each machine.
This paper is concerned with a design methodology for adaptive disturbance observers. One advantage of the proposed design method is that the trade off between the command following and immeasurable disturbance suppression is made transparent in the control system design. This reduces the effort of obtaining a highly accurate system model.
We propose a new technique based upon Genetic Programming to discover new learning rules for Cellular Neural Networks. We choose Genetic Programming because of its ability to discover the values of rule parameter and the form of the rules. A new supervised learning algorithm has been discovered and comparison with other different methods is taken into account
A GA-based approach is presented to derive a reduced model for a high-order interval system. To search for the optimal reduced model, we construct an aggregated error of the Bode envelopes between the original and reduced systems, subsequently minimized by a genetic algorithm. Impulse response energy of the interval systems is obtained to evaluate the performance of the reduced models.
Genetic algorithm using orthogonal design of experiments as the initial population is proposed and evaluated. It improves performance in low variable interaction problems. However, the efficiency is not well recognized in benchmarks of high interactions. The performance of the random initialized and the OD-initialized simple genetic algorithm depend on the problem and the size.