This paper presents a novel gripper capable of actively changing both shape and firmness. The gripper increases its grasp ability by changing its finger posture and firmness suitable for given target objects. In the proposed gripper, each finger is constructed by serially connecting multiple Juzu units. By controlling the angles between neighboring Juzu units individually using two actuators used for sending and bending, arbitrary finger shapes can be generated. In addition, by controlling the tension of the wire that penetrated all Juzu units in each finger, the friction between Juzu units is adjusted and the firmness of finger can be varied. A prototype gripper was designed and developed, and experiments to evaluate the capabilities of changing shape and firmness were conducted. Furthermore, through experiments of preshaping and grasping for various objects with different shape and size, the validity of the proposed method was demonstrated.

To evaluate capacity of the elderly body and the extremities, we developed a measurement system for thigh cross-sectional image. The cross-sectional images were suitable for observing muscle volume. Some previous studies estimated muscular strength based on muscle volume measured the medical images. However there were no studies that used information regarding proportion in human extremities, e. g. muscles, subcutaneous fat and a bone. In this case, the measurements provide the same results even if the proportion is different among two people. We have investigated changes of thigh proportion by age and gender using the cross-sectional ultrasound images. In the survey, 150 ultrasound echo images and the body measurements (height, weight, and so on) were used. We conducted t-test between two groups that have different age, and discussed the causes of their differences. The statistical analysis revealed that the proportion of human thigh was dramatically changed with aging; the shape of thigh outline was deformed and the position of the thigh bone in the cross-sectional image was moved. We considered that the decreasing of the tissue elasticity with aging is responsible for these changes. The proportion analysis can be expected to be a new approach for developing a novel evaluation technique of the elderly body and extremity ability.

To accurately estimate the Beef Marbling Standard(BMS) number of live cattle using ultrasound echo imagings, we have developed a image recognition method by use of a neural network. This paper examines the efficiency of applying Independent Component Analysis(ICA) to the compression of multidimensional image features. ICA can accurately separate a target signal because of its independence assumption, while Principal Component Analysis(PCA), a conventional method, involves decorrelation of the components. We have implemented the estimation tests by use of ultrasound echo imagings of 103 live cattles. Multidimentional texture features extracted from the imagings were compressed by ICA, and then the estimation of BMS number was conducted by using a neural network. The estimation accuracy was evaluated based on the cross validation method. We caluculated the correlation coefficient between the actual and estimated values using 100 different data sets. The results confirmed that the correlation coefficient between the actual and the estimated values was higher by ICA (R = 0.70, p <; 0.01) than by PCA (R = 0.62, p <; 0.01). Also, we conducted the comparison experiments between the ICA based estimation and the estimation by an experienced inspector. The both methods examined the same ultrasound images. Even the experienced inspector failed to estimate BMS number because the estimation requires highly professional skill. The correlation coefficient between the actual and the estimated values was R = 0.70 (p <; 0.01). As a result, we confirmed that the proposed method had much the same capability as the experienced inspector to estimate BMS number.

A control method is presented for a five-fingered artificial hand using EMG signals. The artificial hand is driven by pneumatic actuators and has 15 degrees of freedom. It is difficult to discriminate all the finger motions from just the EMG signals. Therefore, we describe typical hand motions using Petri net and control the finger motions based on this model. The proposed method enables the operator to incrementally control the joints of the five fingers based on the discrimination of the discrete hand motion. The operator only needs to specify the kind of discrete hand motion (i.e., spherical grasp, power grip, hook grip, key grip, and precision grip) and does not need to consider how to control each finger. Each state of the Petri net stores the on/off pattern of the 15 solenoid valves that corresponds to the posture of the five-fingered hand. The hand posture is incrementally varied to complete the desired motion, transitioning to the state in the Petri net based on EMG motion discrimination.We conducted experiments using four able-bodied subjects. In the experiment, the above-mentioned five motions were successfully performed using six-channel EMG signals measured from the forearm of the operator.

Electromyogram (EMG) signals can be measured from human muscles and can be used to anticipate movements. In fact, many researchers have tried to use these signals as an interface tool for a prosthetic hand. However, most of these studies focused on the discrimination performance of the EMG signals, and only discussed the control method for the prosthetic hand. Evaluation of the operating performance while the operator wore the prosthetic hand was seldom reported. This paper proposes a virtual prosthetic control system and presents the analyses of a grasp motion under two different EMG control methods: ON-OFF Digital Control (Digital) and Dynamic Mode Control (DMC). DMC is able to proportionally control the grasping velocity based on the amplitude of the EMG signal. Digital controls the hand at a uniform rate while the amplitude of the EMG signal is greater than a predefined threshold. We conducted experiments with five subjects, and confirmed the usefulness of the developed system.

In this paper, we describe human subject experiments of a multiple-degree-of-freedom (DOF) leg rehabilitation system that we have proposed. The characteristic of the proposed system is employment of a parallel wire mechanism. The rehabilitation devices using a parallel wire mechanism are expected to have the advantages such as system safety and human-friendliness. We have built and are testing a 4-DOF experimental system. The target motions of the experimental system are hip joint flexion/extension, hip joint abduction/adduction, hip joint external/internal rotation, and knee joint flexion/extension. The experimental results showed that the experimental system can carry out 4-DOF motions with different patient postures. Also the results of subjective evaluation showed that the proposed system makes the subjects feel less constrained.

This paper proposes a virtual EMG-prosthetic hand system using a neural network available in an fMRI room. A subject attaches surface electrodes on his forearm, and manipulates the virtual EMG-prosthetic hand visualized with 3D computer graphics by using visual biofeedback projected on a screen in the scan room. The preliminary experiments demonstrate that the intended hand motion by the subject can be determined with high discrimination rates over 96% without decaying fMRI images. Finally, in order to show the validity of the proposed system, we report the neurological experiments that were carried out to analyze human brain functions.

Probabilistic neural networks (PNNs) that incorporate the Bayesian decision rule and statistical models have been widely used for pattern classification. Efficient estimation of the PNN's weights, however, is still a major problem. In this paper, we propose a new training scheme based on a discriminative criterion, maximum mutual information (MMI), and apply this method to the log-linearized Gaussian mixture network (LLGMN) which is one of the PNNs. The MMI training achieves a consistent estimator of network weights, and includes the conventional maximum likelihood (ML) algorithm as a special case. Also, the dynamics of terminal attractor (TA) is introduced for iteration control of the MMI training. Finally, the classification ability of the proposed method is examined with a pattern classification problem of the electromyogram (EMG) signals, and found that the MMI training results in better classification than the conventional ML algorithm.

This paper proposes a new EMG-controlled pointing device using a novel statistical neural network. This device can be used as an interface tool for wearable computers since it does not restrict the operator to being in front of computer devices such as a keyboard or a mouse. The distinctive feature of this device is that we adopt a statistical neural network, which includes a continuous density hidden Markov model, to model the relationship between EMG signals and directions of a pointer movement. The operator can move a pointer in any direction throughout 360 degrees. We also introduced a physical model, such as a mass in a viscous space, into our system to realize a smooth pointer movement corresponding to the operator's force sense. In the experiments, omnidirectional pointer control is achieved using the proposed method and the applicability of our method is confirmed.

Context in time series is one of the most useful and interesting characteristics for machine learning. In some cases, the dynamic characteristic would be the only basis for achieving a possible classification. A novel neural network, which is named "a recurrent log-linearized Gaussian mixture network (R-LLGMN)," is proposed in this paper for classification of time series. The structure of this network is based on a hidden Markov model (HMM), which has been well developed in the area of speech recognition. R-LLGMN can as well be interpreted as an extension of a probabilistic neural network using a log-linearized Gaussian mixture model, in which recurrent connections have been incorporated to make temporal information in use. Some simulation experiments are carried out to compare R-LLGMN with the traditional estimator of HMM as classifiers, and finally, pattern classification experiments for EEG signals are conducted. It is indicated from these experiments that R-LLGMN can successfully classify not only artificial data but real biological data such as EEG signals.

This paper proposes a human-assisting manipulator teleoperated by electromyographic (EMG) signals and arm motions. The proposed method can realize a new master-slave manipulator system that uses no mechanical master controller. A person whose forearm has been amputated can use this manipulator as a personal assistant for desktop work. The control system consists of a hand and wrist control part and an arm control part. The hand and wrist control part selects an active joint in the manipulator's end-effector and controls it based on EMG pattern discrimination. The arm control part measures the position of the operator's wrist joint or the amputated part using a three-dimensional position sensor, and the joint angles of the manipulator's arm, except for the end-effector part, are controlled according to this position, which, in turn, corresponds to the position of the manipulator's joint. These control parts enable the operator to control the manipulator intuitively. The distinctive feature of our system is to use a novel statistical neural network for EMG pattern discrimination. The system can adapt itself to changes of the EMG patterns according to the differences among individuals, different locations of the electrodes, and time variation caused by fatigue or sweat. Our experiments have shown that the developed system could learn and estimate the operator's intended motions with a high degree of accuracy using the EMG signals, and that the manipulator could be controlled smoothly. We also confirmed that our system could assist the amputee in performing desktop work.

Electromyogram (EMG) has often been used as a control signal for a prosthetic arm, which includes information on the operator's motor intentions and the mechanical impedance of joints. Most previous research adopted the control methods of the prosthetic arms based on the EMG pattern discrimination and/or the force estimation from the EMG signals, and did not utilize any knowledge on tasks performed by amputees such as object grasping and soup-spooning tasks. In this paper, a new myoelectric control method is proposed using a statistically organized neural network and an event-driven task model. The task model is represented using a Petri net to describe the task dependent knowledge, which is used to modify the neural network's output. Experimental results show that the use of the task model significantly improves the accuracy of the EMG pattern discrimination.

We propose here a new type of leg rehabilitation system. The aim of the system is to realize the multiple-degree-of-freedom (DOF) training of a leg by manipulating the patient's leg with wires. First, a single-DOF experimental system was built and tested, and the experimental results showed the feasibility of a wire-driven leg rehabilitation system. Next, a 2-DOF experimental system was designed, with a target motion of flexion/extension of both the hip and knee joints. Experiments of single-DOF motion on a test dummy and a male subject showed that the system generated enough accurate motion to support a range of motion exercises. In addition, the results demonstrate that the wire-driven mechanism is user friendly.

This paper proposes a new control method of an externally powered prosthetic hand based on a skeletal muscle model. This method consists of two steps: in the first step, a joint of a prosthetic hand to be driven is selected based on the EMG pattern discrimination using a neural network. Then, muscular contraction levels of flexors and extensors are estimated using other neural networks and used for an impedance control of the prosthetic hand in the second step. In the experiments, natural feeling of the prosthetic control similar to that of a human arm is realized using the neural networks and the impedance control.

Discusses an EMG based control method of a robotic manipulator as an adaptive human supporting system, which consists of an arm control part, a hand and wrist control part and a graphical feedback display. The arm control part controls joint angles of the arm according to the position of the operator's wrist joint measured by a 3D position sensor. The hand and wrist control part selects an active joint out of four joint degrees of freedom and controls if using an impedance model based on the EMG signals. A distinctive feature of our method is to use a statistical neural network for EMG pattern discrimination. This network can adapt to changes of the EMG patterns according to differences among individuals different locations of the electrodes, time variation caused by fatigue or sweat, and so on. It is shown from the experiments that the hand and wrist motions can be controlled based on the EMG signals sufficiently. It may be useful as an assistive device for a handicapped person.

This paper proposes a training system based on EMG signals for prosthetic control and the development of its prototype. This system aims to enhance three kinds of control ability: muscular contraction, cooperation among several muscles, and the timing of EMG generation. For EMG signal processing a statistical neural network is used which can adapt itself to changing EMG patterns according to the differences among individuals, the different locations of the electrodes, the time variation caused by fatigue or sweat, and so on. During training, EMG signal information is displayed in the feedback monitor. The experiments have been conducted using the prototype system. The subject is a 51 year old man who had his forearm amputated when he was 18 years old. After training for five days, the ability to manipulate the EMG signal of the subject has been enhanced and the effectiveness of this system is shown.

Proposes a new probabilistic neural network (NN) that can estimate the a-posteriori probability for a pattern classification problem. The structure of the proposed network is based on a statistical model composed by a mixture of log-linearized Gaussian components. However, the forward calculation and the backward learning rule can be defined in the same manner as the error backpropagation NN. In this paper, the proposed network is applied to the electroencephalogram (EEG) pattern classification problem. In the experiments described, two types of a photic stimulation, which are caused by eye opening/closing and artificial light, are used to collect the data to be classified. It is shown that the EEG signals can be classified successfully and that the classification rates change depending on the amount of training data and the dimension of the feature vectors.

This paper proposes the concept of a human-robot interface as rehabilitation aid and develops the prototype system. The prototype system aims to be used as a controller for the robotic manipulator and as rehabilitation system for the handicapped person. In order to adapt the system to the characteristics of the operator's electromyogram (EMG) signal, the EMG pattern discrimination method using the neural network is utilized as an essential technique of our system. In the experiments, it can be seen that the robotic manipulator can be controlled with high accuracy using the operator's EMG signal, and that the adaptive learning of the neural network improves the discrimination ability of the EMG signal. The rehabilitation program and biofeedback are also discussed.