Shinzo Kitamura

Particle-size distribution determination using optical sensing and neural networks

We present an inverse technique to determine particle-size distributions by training a layered perception neural network with optical backscattering measurements at three wavelengths. An advantage of this approach is that, even though the training may take a long time, once the neural network is trained the inverse problem of obtaining size distributions can be solved speedily and efficiently.

Numerical simulation of the retrieval of aerosol size distribution from multiwavelength laser radar measurements

A numerical investigation was carried out into the feasibility of deriving the aerosol size distribution from aerosol volume extinction and backscattering coefficient measurements by a multiwavelength laser radar. This study employs the regularization method for matrix inversion with the first-order B-spline function as basis functions to approximate the aerosol size distribution. The results of numerical simulations show that (1) the effects of roundoff errors in the numerical calculation are negligible and the approximation errors in the size distribution by the B-spline function are small, (2) the reconstruction errors in the size distribution at its peak are about twice as large as the relative measurement errors when the Lagrange multiplier, which determines the degree of smoothness in the reconstruction, is suitably chosen, and (3) the variation in the complex refractive index due to the humidity change does not produce large errors in the size distribution.

Autonomous trajectory generation of a biped locomotive robot

Introduces a hierarchical structure for motion planning and learning control of a biped locomotive robot. In this system, trajectories are obtained for a robots joints on a flat surface by an inverted pendulum equation and a Hopfield type neural network. The former equation is simulated for the motion of the center of gravity of the robot and the network is used for solving the inverse kinematics. A multi-layered neural networks is also used for training, walking modes by compensating for the difference between the inverted pendulum model and the robot. Simulation results show the effectiveness of the proposed method to generate various walking patterns. Next, the authors improved the system to let the robot walk on stairs. They set up two phases as a walking mode; a single-support phase and a double-support phase. Combination of these two phases yields a successful trajectory generation for the robots walking on a rough surface such as stairs.<<ETX>>

Q-Learning with adaptive state segmentation (QLASS)

Q-learning is an efficient algorithm to acquire adaptive behavior of the robot without a priori knowledge of the sensor space and the task. However, there is a problem in applying the Q-learning to the task in the real world-how to construct the state space suitable for the Q-learning without knowledge of the sensor space? In this paper we propose Q-learning with adaptive state segmentation (QLASS). QLASS provides a method to segment the sensor space incrementally, based on sensor vectors and reinforcement signals. Experimental results show that QLASS can segment the sensor space effectively to accomplish the task. Furthermore, we show the obtained state space reveals the fitness landscape.

Autonomous trajectory generation of a biped locomotive robot using neuro oscillator

The trajectory of a biped locomotive robot is generated using a neuro-oscillator. This oscillator consists of four neuron cells which are mutually coupled with inhibitory connections. This model shows a stationary periodic oscillation for an appropriate set of parameters. Stability analysis of the neuro-oscillator can be performed by the linearization method. A stationary periodic oscillation appears in the unstable region for equilibrium states, and this periodic oscillation generates a trajectory for stationary walking by assigning the state variables of neuron cells to joint angles of the robot. Simulation studies confirm the relevancy of the proposed method.<<ETX>>

Motion generation of a biped locomotive robot using an inverted pendulum model and neural networks

The authors introduce a hierarchical structure for motion planning and learning control of a biped locomotive robot. The motion of the center of gravity of the robot is simulated by that of an inverted pendulum. A Hopfield-type neural network is used for solving the inverse kinematics in order to obtain joint positions from the position of the center of gravity and the position of the toes calculated from the equation of an inverted pendulum. A feedforward input, generated by a three-layered neural network, is used as a correcting reference input to make the motion of the center of gravity follow that of the inverted pendulum. Simulation results showed that stationary walking was successfully achieved. The proposed method thus provides an autonomous motion generation where only the position and velocity of the center of gravity of the robot for each step are given a priori.<<ETX>>

Modeling and genetic solution of a class of flexible job shop scheduling problems

We consider an extended class of flexible job shop scheduling problems. First, we translate the problem into a mathematical programming formula, i.e., a mixed-integer programming problem. This makes it possible to apply standard packages of mixed integer programming solvers and, while lots of computational time is required in general, to obtain the optimal schedule. Then, in order to seek the schedules close to the optimal for larger-scale problems, we newly design a solution method by adopting genetic algorithms based on the formula. Through some computational experiments, the effectiveness and the possibility of the proposed approach are examined.

An application of branch-and-bound method to deterministic optimization model of elevator operation problems

In this paper, we propose a framework for obtaining the optimal solution of an elevator operation problem by applying branch-and-bound method, where it is assumed that all information about the passengers are given. The problem is solved by determining the assignments of passengers to elevators and the processing order of passengers for each elevator. The validity of an existing rule to decide a car service is examined by comparing the results with the optimal one.

A continuous age model of genetic algorithms applicable to optimization problems with uncertainties

We study a method of optimum seeking in an uncertain environment by extending the conventional genetic algorithms. So far, we have extended genetic algorithms by introducing an age structure, where the key point is to evaluate an individual not directly by an objective value of a corresponding solution currently observed, but by accumulating values which have been observed at preceding generations. We newly introduce a continuous age model of genetic algorithms to accumulate the values adequately. Then, the effectiveness of the proposed method is investigated through some computational experiments. As a result, it has been shown that the stability is higher and the possibility of falling into local optimum is lower in using the new age model than in using the former one.

A hybrid neural network system for the rainfall estimation using satellite imagery

Hybrid neural networks composed of a self-organizing map (SOM) and three-layered feedforward neural networks have been developed and applied for rainfall estimation using satellite imagery. The SOM classifies an input vector extracted from satellite imagery, then one of the feedforward neural networks is chosen according to the class to give the rainfall estimation. In order to train the hybrid neural network, adjoining seas of Japan were selected as testing area. Hourly GMS infrared imagery data and simultaneous ground truth data (the area average of rainfall observations and radar/raingage composite data) were collected from AIP/l/sup 2/ data sets. The SOM is trained to classify the textural feature vectors extracted from the imagery data, and tuned by learning vector quantization method. The feedforward neural networks are trained to give the estimation by back propagation algorithm. Fairly good correlation coefficients about 0.8 are obtained between the estimation and corresponding ground truth for the unlearned test set. Furthermore, SOM with a recurrent structure for processing the temporal information has been proposed and tested.