Hajime Kanada

A Solution of Inverse Kinematics of Robot Arm Using Network Inversion

The network inversion solves inverse problems using a multilayer neural network. The inverse kinematics to estimate the joint angles of the robot arm from the end effectors coordinate is an inverse problem. In this study, the network inversion is applied to the inverse kinematics that estimates joint angles of robot arm. The inverse problems are often difficult to solve because of the ill-posedness. The regularization method is necessary to solve the ill-posed inverse problem. In this study, the regularization method is examined to solve the ill-posedness

Time-frequency analysis of impact sound of composite materials

The method was proposed for examining the characteristics of materials from impact sound. This method was designed to estimate the elasticity of materials, nondestructively and easily. It becomes difficult to estimate accurate elastic moduli of the vibration as the temperature rises, because of the noise and damping. In this study, we introduce the time-frequency analysis for estimating to estimate the accurate elastic moduli. We examine the effectiveness of the short-time Fourier transform and the wavelet transform.

Solution for Ill-Posed Inverse Kinematics of Robot Arm by Network Inversion

In the context of controlling a robot arm with multiple joints, the method of estimating the joint angles from the given end-effector coordinates is called inverse kinematics, which is a type of inverse problems. Network inversion has been proposed as a method for solving inverse problems by using a multilayer neural network. In this paper, network inversion is introduced as a method to solve the inverse kinematics problem of a robot arm with multiple joints, where the joint angles are estimated from the given end-effector coordinates. In general, inverse problems are affected by ill-posedness, which implies that the existence, uniqueness, and stability of their solutions are not guaranteed. In this paper, we show the effectiveness of applying network inversion with regularization, by which ill-posedness can be reduced, to the ill-posed inverse kinematics of an actual robot arm with multiple joints.

Neural network localization of a steel ball in impact perforation images

The estimation of characteristics from the impact perforation process of the material by the high-speed photograph system has been studied. In this method, the characteristic of the material is estimated from the continuous images after the steel ball perforates into the material specimen. In this study, we propose to use the neural network to localize the steel ball in the continuous images.

Bispectrum analysis for impact sound of composite materials

A method to estimate the elastic moduli of a composite material by the sound that occurs at an impact on the composite material was proposed in the field of the material engineering. To estimate elastic moduli from the impact sound, we have to estimate the natural frequencies of the vibration of the material from the obtained time-series waveform. In this report, we show that the natural frequencies of the vibration of the composite material can be estimated by the bispectrum analysis more clearly than the power spectrum analysis.

A neural network model for realizing geometric illusions based on acute-angled expansion

Recently, the study of geometric illusions has been remarkable in the field of artificial intelligence and computer vision etc., to make the spatial recognition ability of humans clearer and to apply it technologically. In the technological field, it is often supposed that the visual illusion takes place at the retina level to explain the phenomenon by the simple and unified model. However, the possibility of the illusion occurring at the higher visual field, which has been shown by various psychological and physiological studies, cannot be ignored. Recently, the acute-angled expansion recognizing the angle within the crossing lines being different from the actual angle has been reported by psychological studies. It is explained by the lateral inhibition among the orientational neurons in the cerebrum cortex. It is interesting to explain the higher visual field participating in the geometric illusion. In this study, we compose the neural network model that realizes the acute-angled expansion, by lateral inhibition among the orientational cells on the higher visual field. Moreover, we test the model by computer simulation on the intersecting line segments.

Network inversion for complex-valued neural networks

The network inversion method using multilayer neural network was proposed and has been studied for the purpose of solving inverse problems. The original network inversion method has been applied to usual multilayer neural network with real-valued inputs and outputs. The method using neural network with complex-valued inputs and outputs is needed for the general inverse problems including complex numbers. In this study, the complex-valued network inversion method is proposed to solve the inverse problems with complex numbers. A simple simulation of inverse mapping is carried out to confirm the proposed method

Regularization for complex-valued network inversion

Network inversion solves inverse problems to estimate cause from result using a multilayer neural network. The original network inversion has been applied to usual multilayer neural network with real-valued inputs and outputs. The solution by a neural network with complex-valued inputs and outputs is necessary for the general inverse problems including complex numbers. The complex-valued network inversion method has been proposed to solve the inverse problems with complex numbers. In general, there is a problem attributable to the ill-posedness on the inverse problems. To solve the ill-posedness, the regularization is used to add some conditions on the solution. In this study, we propose to introduce the regularization to the complex-valued network inversion.

Inverse estimation of joint angles of robot arm by network inversion

The inverse problems that estimate the cause from the results have been studied in the various engineering fields. The network inversion was proposed for solving inverse problems using a multilayer neural network. In this study, the network inversion is applied to the inverse estimation of the joint angles of the robot arm. The effectiveness of the proposed method is shown in the simulation of the three degree of freedom robot arm.

Impact Perforation Image Processing Using a Neural Network

The evaluation of materials characteristics from the impact perforation images has been studied in the material engineering fields. In this method, the steel ball is shot into the material specimen, and the characteristic of the material is estimated from the steel balls behavior. However, the observation of steel balls behavior is often difficult because of the scattered fragments of the specimen. We have proposed to use the neural network to estimate the steel ball position in the impact perforation image. However, the miss-recognition of the steel ball was often seen because of the influence on the scattered fragments of the specimen. In this study, the preprocessing of the image with the high-pass filter is introduced to improve the performance of the recognition of the steel ball. We examine two types of filters using the Hanning window and the Blackman window