In our laboratory, we aim at the realization of the Upper-Limb power assistance system based on Electroencephalogram(EEG). In this study, we focused on potential fluctuation occurred before and after voluntary movement called Movement-Related Cortical Potential (MRCP) and investigated features of elbow, shoulder and coordinated movement. As a result, P+160 which is one of the MRCP components at C3 of elbow movement is bigger than shoulder movement, and compared to single movement, the potential of coordinated movement fluctuated characteristically in all the subjects. The results suggest that it is possible to recognize each movement by using these features.

In recent years, neural network (NN) has made excellent outcomes on the problems that are difficult to analytically solve due to their complexity. NN has a critical problem, called catastrophic forgetting: memories for tasks already learned would be broken when a new task is learned additionally. This problem interferes with continuous learning required for autonomous robots. It is supposed that the catastrophic forgetting is caused by rewriting the whole network by backpropagation of learning signals. This study therefore propose a method to mitigate the catastrophic forgetting using a model that learns only the output layer of recurrent NN, called reservoir computing. Instead of giving the network randomly like the conventional reservoir computing, the network structure and its weights are designed based on a fractal complex network. This network modularized memories for multiple tasks and mitigated the catastrophic forgetting.

The sweeping robot plans path and moves along its prior path. However, the outdoor sweeping robot cannot move as expected. The uncertain movement of the robot is caused by some environmental disturbance such as unsmooth loose soil and wind. For this reason, we consider path planning considering robot uncertainty because of environmental disturbance. This study considers only wind blow as a disturbance. In this study, we define the objective function by investigating the relation between the environmental effect and the prior path based on transition probability model.

Soft-bodied robot with ambiguous boundary between driving part and body such as gel robot is expected to give action that cannot be realized by conventional robot composed of metal parts in the field of soft robotics. This paper discusses a motion pattern of soft-bodied agent by using one-dimensional cellular automaton through an elastic robot simulation by using voxel model. We propose a classification approach of soft-bodied robot motion by using a one-dimensional cellular automaton (CA). We give a volumetric actuator arrangement by CA in the body. As the result, we indicate that an ordered class and a chaos class give the chaotic behavior to the agent without an external noise, and we discuss that the chaos behavior is dominated by the positional deviation of actuator arrangement.

There are many studies on neural networks to control autonomous robots operating in real world environments. The well-trained neural networks give good solutions in many cases. However, we cannot understand the reason why the solutions are appropriate because the neural networks do not give any explanations just like black box does not. This incomprehensibility excludes neural networks in some applications that require the exact proofs of propriety of outputs such as autonomous car driving and medical operations. We propose a sort of hierarchical neural networks with a twodimensional internal layer to visualize the learning results and to provide intuitive understanding on the input-output relations. Some experimental results on the obstacle avoidance learning with real robot are also shown in this paper.

This paper proposes a planning method of optimum picking point for piled flexible pouched objects by vacuum suction robotic hand. The bin picking problem for the deformed and multiple stacked works is not so easy because of the deformation. A combinational method of model based geometrical estimation of works and stochastic planning based Bayes’ estimation is proposed. Estimation results of proposed method and machine learning are compared. The total method is verified for bin picking experiments of multiple deformable food pouches by using a parallel link robot and a suction pad.

Among the cardiac nuclear medicine examination with the target tissue as the heart, in the examination using the 99m-Tc formulation as the radiopharmaceutical, there is a tendency to accumulate in the liver or the gall bladder other than the heart, which adversely affects the image of the heart, it interferes with diagnosis. Usually, for this problem, it is necessary to prepare an image for diagnosis after manipulating the image manually and performing pretreatment such as removal of the other than cardiac accumulation portion, but a heavy burden is imposed on it. In this study, we believe that if we can use Deep Learning to remove accumulation other than the heart at the stage before image reconstruction, it will lead to reduction of artifacts (virtual images) in image diagnosis. We construct a cardiac detection system aiming at creating a transverse section from which outside of cardiac accumulation is removed by using Convolutional Neural Network and Regions with CNN features.

This paper described the grasping method of flexible objects using a biomimetic octopus vacuum gripper with a water-filled pad. The pouch containers of flexible objects change the shape during lifting. The previous biomimetic octopus gripper cannot follow the shape of pouch containers, because the gripper harden in grasping. The previous gripper cannot grasp the light containers with a weight of 11 g. We proposed the flexible water-filled pad. The pad can follow the change in shape of objects. The fabricated gripper has the ring-shaped water-filled pad. This gripper can grasp the liquid soup filled pouch containers of weight 470 g.

The variable stiffness suction pad can deform around the object. A few researches showed the grasping the rigid objects using the pad. The flexible pads became rigid on vacuum suction. We proposed the relationship between suction sequence and suction performance. We tested five suction sequences; the stiffness of pad is 1) harden before press on object, 2) harden after press, 3) harden with suction, 4) harden after suction, and, 5) flexible during operation. When the fabricated pad was the low stiffness in suction, the pad can grasp the uneven metal objects. The suction force of the sequence 5) was lowest in five suction sequences. In the sequence 3) and 4), the grasping performance was almost same. We have shown that the variable stiffness suction pad has the suitable suction sequence for the shape of object.

Recently, many human works are automated by robots. In addition, due to lack of workers in pottery companies, automation of transfer paper attachment work on a pottery's surface is needed. The work is performed manually by human. Therefore, the quality of products changes due to differences in technical skills. It aims to improve accuracy and productivity by using robots. In this paper, we develop devices that automatically attach the transfer paper.

Introduction of robots is progressing for the purpose of reducing work load and enhancing productivity in industrial fields where there are many mechanical work. In addition, it is expected that introducing robots that work in place of humans in extreme conditions such as space and disaster sites. To improve the dexterity of industrial robots, we investigate robotic assembly of fixtures used in the production lines. We not use a parallel gripper or dedicated hand, but use a 4-finger 12 d.o.f. hand with a camera attached on the palm and a 6-axis force sensor on the fingertip. In this paper, we investigate assembling fixture with bolts. The fixture used this time can't be assembled by the hand alone because the counter bore is done. Accordingly we grasp the tool made with 3D printer by the hand and tighten bolts. Also, in order to improve the accuracy of bolt insertion and reduce the time, we improve the fixture installation position correction method. We use a camera for object recognition and use a fingertip 6-axis force sensors to check the tightening torque and reliably grasp.

The development of active-caster and its dead reckoning system are presented in this paper. We designed and built a new type of the active-caster which is an omnidirectional mechanism for the industrial robots. In addition, pulse motors are used for precise movement of active-caster. For confirming the motion of the prototype active-caster, we developed the dead reckoning system. This system is constructed by the kinetic models of the active-caster for prototype robot with two activecasters. In experiments, robot 3D trajectories were tested by using this system and 3D measurement system Quick Mag. Some experiments were conducted to evaluate movement accuracy of active-caster in two types of motors. Also we confirmed validity of the proposed system.

This paper proposes a method of regrasping objects using a dual-arm robot with general-purpose hands. For regrasping a target object with the general-purpose hand, error of grasping have to be considered. The proposed method performs positioning the target object at optimal posture for regrasping using visual servoing approach to reduce effect by the error of grasping. The proposed method computes the optimal posture for grasping by maximizing minimum singular values of image Jacobian of candidate postures for the target object. Experimental results show that the proposed method achieves regrasping the target object with the general-purpose hands and performs positioning the target object with less than 0.8[mm] position accuracy.

In this research, we propose multi-class image classification method based on EEG signal using stacked autoencoder, which is one of the deep learning algorithms, in order to obtain high classification rate for multiple thoughts. We measure brain wave signals with respect to four different thoughts, process measured data based on fast fourier transformation, and train stacked autoencoder by using the data. Then the trained model is verified by evaluation data.

The purpose of this research is to develop common EEG classifiers using EEG data measured from plural subjects. In this paper, personalized classifiers and common classifiers to classify two kinds of images are created and evaluated using Neural Network and CNN which is one of the Deep Learning Methods. As a result, both of personalized classifier and common classifiers show almost same performance.

The purpose of our research is to develop a grasping training device which effectively encourages post-stroke patient to recover functional grasping motion of a paralyzed hand. Our previous research reported that our grasping training device successfully supported a rehabilitation process of a post-stroke patient to regain a grasping motion by using a grasping object. It however needs a patient voluntary motion to fix his wrist and thumb on a “hand rest” during the rehabilitation. In this study, a simple mechanism to move a wrist joint and carpometacarpal(CM) joint of a thumb is installed so that wrist joint flexion and CM joint flexion could be supported throughout rehabilitation, synchronizing grasping motion of four fingers. We verify that the developed device can apply force evenly to user's ball of fingers in grasping motion.

In this research, we focus on the relationship between electroencephalography (EEG) signals and torques generated by elbow or/and shoulder joint when performing flexion and extension, and aim to succeeded in the power augmentation by estimating torque using EEG signals for the multi-DOFs wearable upper-limb exoskeleton robot developed in our laboratory. In this paper, we propose a method to distinguish motion intention of elbow or/and shoulder joint performing extension and flexion by using EEG signals when wearing the upper-limb exoskeleton robot. The proposed approach is verified by experiments, and the results suggest that motion intention distinguishment of elbow or/and shoulder joint performing extension and flexion based on EEG signals is feasible, and demonstrate the potential of using EEG signals via brain-machine interface to support human activities.

In this paper, in order to enhance the identification accuracy of grasping behavior via electroencephalogram (EEG), we attempted to solve the problem of indeterminacy of the estimated EEG, which is a drawback of independent component analysis (ICA) used to restore the EEG before mixture from the measured EEG, and we also attempted to increase the number of electrodes for EEG measurement. Specifically, the separating matrix in ICA was determined in advance through preliminary experiments and was fixed uniquely, and the number of electrodes was increased from 2 to 6. Then, identification experiments were conducted, in which existence or nonexistence of the grasping behavior and right grasping or left grasping were identified. As a result of the experiments, the identification accuracy was improved. Thus, the validity of fixing the separating matrix and increasing the number of electrodes was verified.

Recently, many researchers are focusing on Brain Computer Interface (BCI) and Brain Machine Interface (BMI) as a new method for controlling machines, in which EEG signal is used as a control signal. In this paper, a BCI system we constructed is introduced. We use an inexpensive commercially available product to obtain brain wave data. In addition, we built a motor control system, which shares the obtained brain wave data with the computer for measuring the brain wave. We analyzed the brain wave using wavelet transform. The EEG data is communicated via Bluetooth to a PC, in which wavelet analysis is done. Then analyzed data are sent to Raspberry Pi 3 using file sharing system Samba. We conducted an experiment and confirmed that the rotating direction of the DC motor was successfully controlled based on the analyzed EEG data.

In policy search methods, importance sampling is widely used to reutilize samples drawn from previous sampling distributions that are usually different from the current one. Previous studies create uniform mixtures of previous sampling distributions as proposal distribution. To further improve sample efficiency, we introduce adaptive multiple importance sampling that optimizes the mixing coefficients to minimize the variance of the importance sampling estimator. We apply the proposed method to several policy search methods and experimental results on some benchmark control tasks show that all the methods improve sample efficiency.

In recent years, the input interface has evolved diversely, and the inputs to the interface varies from gestures to voices. However, such an interface is predefined for input and output by the designer, and it is not known whether the interface moves as intended by the user. In addition, the interface tends to lose the intuitiveness, as the degree of freedom of the controlled object increases, and to be complicated. It is a big challenge for designers to create an intuitive and easy-to-use interface for all users. In this research, we extract human intention from body movements, especially hand movements, and associate with movements of multi-degree of freedom spider robot, thereby constructing interface to operate robot only from hand movements. Since the hand movements reflects the intention that varies depending on the individual user, the correspondence between the hand movements and the operation of the controlled object is built using machine learning. This aims to create an interface optimized for individual users. The interface constructed in this way had difficulty in operation performance in one subject, but showed sufficient operation performance in other subjects.

In this study, we developed a method “Statistical Q-learning” a combination of statistical learning and reinforcement learning (Q-learning), which adjusts sub reward by itself using statistical information to fasten the learning speed of Q-learning. The experiment result of learning swing action shows that Statistical Q-learning can learn many stated problems, which is difficult for conventional method to learn stably.

In recent years, the development of robots for substituting and supporting human work in an environment, where we live is actively performed. These robots need to behave with high versatility. So, a method called machine learning is effective to experience these behaviors. For robots to learn flexible motions like human, a method called deep reinforcement learning is generally used. But learning complex motions requires a lot of time, and improvement of learning efficiency is necessary. Therefore, we propose a method which combines deep reinforcement learning and imitation learning that imitates the behavior of the experts. Manipulators acquire a certain level of motion by imitation learning, and then refine its motion by deep reinforcement learning. The process makes learning complex motions efficiently. Experiments of opening a door with a manipulator shows the usefulness of this method.

Smartphone based robotic platform are built to understand the reward and affective systems of biological creatures through the combination of learning and embodied evolutionary methods. The robots should share the same intrinsic constraints as natural agents, such as foraging and mating. In this paper, we introduce the smartphone-mounted two wheels inverted pendulum mobile robot hardware and their real-world environment for acquiring such behaviors. We use a policy-based reinforcement learning method called EM-based Policy Hyperparameter Exploration with Adaptive Baseline (EPHE-AB) for the agents to learn vision-based foraging and mating. The results show the feasibilities and indicate a potential in further studies.

In recent years, as the place of active autonomous mobile robot moves to human society, it is necessary to operate while anticipating the movement of a person in a situation such as a crowd. Conventionally, this prediction is often performed under the assumption that the speed of the target object is calculated, and the speed is maintained and acted. Therefore, it is impossible to respond to actions with changing speed from the relationship with the surroundings such as a person passing by. Therefore, in this study, motion prediction of dynamic obstacles is performed using 3D-CNN. 3D-CNN is used for time series 3D point cloud to extract shape information and time change. Based on the relationship with the surroundings of each extracted information, we estimate the future position of each point cloud. By doing this, it is possible to predict motion of the object in a situation difficult with the conventional method.

Laparoscopic surgery, which is also called as a minimally invasive surgery, is a surgical technique that is associated with an accelerated post-operative recovery. However, it can only be performed by surgeons possessing advanced surgical skills. One of the main challenges in laparoscopic surgery is the restriction of the free motion of forceps because of the limited degrees of freedom imposed by a trocar. Recently, to overcome this problem, many master–slave manipulators with articulated forceps have been used in laparoscopic surgery. The wrist mechanism of the articulated forceps affects the controllability and range of motion of a slave manipulator in the abdominal cavity of a patient. Therefore, it is important to improve the wrist mechanism of articulated forceps in order to perform robot-assisted laparoscopic surgery. This study proposes a new wrist mechanism for the usage of articulated forceps in laparoscopic surgery. The degrees of freedom of the proposed design are provided by three motor-driven axes that use various wires and pulleys (pitch, yaw, and gripper axes). The kinematic model of this mechanism is decoupled between the pitch axis and yaw axis using a very simple mechanism with arc-shaped guides and wire guide holes. Furthermore, the effectiveness of the mechanism is demonstrated using various tests on a prototype.

The operation of the heart-lung machine at the medical site is a heavy burden on the medical staff. In this paper, we report about an attempt to develop an occluder which automatically controls the flow rate of the blood removal circuit in order to reduce the burden.

Among those suffering from diabetes, measuring the pressure between the shoe and the foot to prescribe appropriate shoes to those suffering from "diabetic foot leg lesion" whose foot shape changes can be measured I made it. By attaching a urethane resin imitating a hallux valgus to the thenar eminence, it was observed with the measuring instrument what influence would be caused by the presence or absence of urethane resin.

Recently, the lack of expert doctor for endoscopic surgery has been recognized as one of the medical issues in Japan. In order to cultivate the expert doctors, the training environments for endoscopic surgery have been developed and widely used all over the world. However, neither the functions of correcting wrong operation nor instructing collect operation for each situation have been installed. Then, this study aims to develop a VR training system, where a trainee can learn the operation skill of an advising doctor. This paper briefly introduces the system we have developed and shows the experimental results.

In this report, we proposed a method of classification of fascia and other tissue using Histograms of Oriented Gradients (HOG) and Support Vector Machine (SVM) in ultrasound images. The muscle layer can be determined by detecting the fascia. The decline of muscle and bone density due to aging is increased the likelihood of injury. So it is considered a serious problem. To cope with this problem, we proposed a method of automatic fascia classification to visualize muscle thickness. Our method use SVM based on the texture analysis of ultrasound images. Our method achieves about 90% Accuracy and Recall by considering that the fascia is a continuous tissue. Experimental results show the effectiveness of our proposed automatic fascia extraction method.

Ultrasound has a lot of advantages in observing diseased organ areas compared with other imaging diagnostic modalities. Nevertheless, the presence of acoustic shadows in ultrasound images can impede observation of the diseased area, and although the ultrasound probe can be moved to scan for obtaining ultrasound images that avoid the bones, this limits the current imaging field of view and viewpoint which is favorable to observe the affected area. To cope with these problems, we have designed a conceptual model with the aim of expanding the imaging range for organ diseased areas by compositing the images to complement areas with acoustic shadows. In this report, we propose a method for extracting the composited acoustic shadow element from the ultrasound images of kidneys on the basis of the principles of ultrasound image generation and qualities of bone-generated acoustic shadows. Experimental results show our proposed method extracts acoustic shadows with high accuracy.

Collaboration between human and robot such as object assembly in real environment is expected. Since the difficulty of realizing human-level intelligence, developing collaborative robots as “human support machine” is realistic. To realize it, hierarchical (long-term and short-term) goal planning and replanning ability is necessary. In this research, we develop a learning-based system that enables robots to realize planning and replanning ability for long/short-term goals. Experimental results demonstrate that a robot with the developed system successfully collaborated with human and achieved task goals by using hierarchical planning ability. Even the situation was changed, the robot was able to generate replanned movement to adapt to the change.

In this study, we propose, as a method of exoskeleton robot control, to estimate the voluntary lower limb muscle activities that are lost after spinal cord injury (SCI). This method is based on spinal cord mapping of the remaining muscle activities and its online matching to the one obtained from healthy participants, taking into account the muscle synergy of the whole body during walking motion. Here we report on the analysis of the synergy of whole body muscle activity during walking and its spinal cord mapping using Non-negative matrix factorization(NMF), and also on the computation of transformation matrix to estimate intended lower limb muscle activity from the remaining spinal cord activity. Implementation of the proposed method using the right leg of HAL (Hybrid Assistive Limb) and walking experiments with a healthy participant are also reported.

Body representation is one of bases of adaptive human behavior, e.g., postural and motion control, tool use, imitation and self-recognition. However, it is still difficult to understand how we acquire and apply it to such behavior. Here, we propose a novel approach to explain the mechanism of acquisition and application of body representation from the control engineering viewpoint, based on analytical expression of neural networks which approximate forward dynamics. In this paper, it was suggested by simulating 2-link planar manipulator control in various situation, that body representation acquired through the proposed method had plasticity and was able to be applied to manipulator control, state estimation and prediction.

Exploration, which is important for sensorimotor development, is said to be realized by intrinsic motivation. Intrinsic motivation is “the doing of an activity for its inherent satisfactions rather than for some separable consequence”. It is not clear how exploration is driven by intrinsic motivation. In this study, we conducted an experiment comparing behaviors with or without intrinsic motivation using a computational model of intrinsic motivation. As a result, intrinsic motivation drives the exploration to emerge more posture and action and the exploration to touch environment continuously. In addition, an agent with intrinsic motivation reproduces behaviors with high coordination, which is one aspect of early sensorimotor development.

While avoiding ethical problems, it is necessary to construct integrated simulator of the central nervous model and musculoskeletal model in order to investigate the relationship between the activity of human brain and human motion. We construct this integrated simulator by using mouse brain model for a brain structured model and integrated simulator of spinal nervous model and human musculoskeletal model. We implement the lateral corticospinal tract and the posterior column-medial lemniscus pathway for 64 muscles out of 374 muscles in human whole body. We success to construct the integrated simulator that can generate motion of a specific part of the human musculoskeletal model by firing of a specific area of the mouse brain model.

The use of robots is expected not only for industrial fields but also for non-industrial fields, such as medical, welfare, disaster rescue, and so on. Non-industrial robots, however, are always exposed to a change of working environment to be different from industrial robots. So they are required to move with taking a various pose to be adaptively desired for a change of working environment.

In this paper, we aim to apply motion generation method using machine learning based on human demonstration data to generation of task motion of home robot. A generation model by using VAE-GAN based on image data at task from multiple viewpoints is created, and a network model that can recognize the positional relationship between the robot hand and the object is also created. In addition, LSTM is learned by inputting latent variables generated by the image network model and joint angle data of the robot at each time, and network model capable of outputting the joint angle command value necessary for the task is created. Pushing task of the chair by the home robot is executed on the dynamics simulator by the learned network model and its usefulness is discussed.

The shoulder and wrist of the robots have multi-degrees-of-freedom joint. The parallel mechanism using three actuators is applied to this joint using the spherical motion mechanism. This study proposes a shoulder joint that realizes the motion of three degree of freedom by using of the parallel mechanisms that consist of three serial links manipulators. This paper deals with the proposed the kinematics of the parallel link mechanism composed of the kinematics of three serial links manipulator. Perform motion analysis of the proposed parallel link mechanism focusing on characteristics of spherical motion mechanism

Free Energy Principle enables agents to understand the generative models of the environment, to have beliefs about their current states by perceptual inference, and to behave adaptively to environmental changes by minimizing their prediction errors. This work combines Free Energy Principle from computational cognitive neuroscience and Deep Learning from computer science, suggesting its potentials to be applied to the understanding of agents' adaptive behaviors in complex environments. As an example, this paper shows that an agent can behave adaptively when it is given an expert's goal-directed belief.

In this report we propose a novel method for shape extraction and modeling of MRI prostate contours. Transperineal targeted biopsy with real-time fusion image of multiparametric magnetic resonance image and transrectal ultrasound image have a problem that the diagnostic ability is uneven in accordance with the skill of the medical professionals. To cope with these problems, we propose a method, which is not affected by diagnostic ability, to extract prostate contour utilizing deformable superellipse models. The characteristic of oue method is to implement the region division function for the deformable superellips models to enhance the expressiveness to handle the left and right asymmetry of the prostate contours. Experimental results show our proposing method is effectiveness.

The existing head tracking systems for the endoscope holding robot have problems of restricting the operator's movement or interfering with the operator. This research proposes a system that can measure the head motion by simply putting a device on the head. Using the system, we also constructed an operating system for the endoscope holding robot. The image obtained from the camera on the head is wirelessly transmitted to the external PC, and the ORB-SLAM is executed on the PC to compute the angular velocity and the translation velocity of the head. For the developed system, we evaluate the dynamic characteristics and the result shows the delay of the measurement was less than 50 ms within the range of actual operation, which is considered that this delay does not affect the operation. We also carried out an in vivo experiment using a pig and show the effectiveness of the developed system.

Recently, single-port surgery (SPS) has been performed frequently in the field of laparoscopic surgery because its invasiveness is low compared with the conventional laparoscopic surgery. However, surgeons need considerable skill to perform SPS. In our previous study, a haptic surgical robot for SPS was developed, which is composed of 2 forceps robots and 1 laparoscope robot. In this paper, to expand a view field of the laparoscope robot, a tip bendable laparoscope manipulator was developed. In addition, operation method of zoom-in and zoom-out of the laparoscopic image was improved so that the laparoscope robot could be operated more intuitively. Experimental works were carried out to verify an effectiveness of the developed tip bendable laparoscope manipulator and the improvement of operation method of the laparoscope robot.

There is a surgical method called CT-guided Interventional Radiology(IVR). We have been developing surgery support robot for CT-guided IVR. When this robot is used, there is a targeting phase in which the position of needle is navigated to the initial position before insertion. Currently targeting is done by doctor's manual operation. However, manual targeting is difficult. In order to solve this problem, we implemented a method of automatic targeting. However, due to assembling errors and mechanical stiffness, the accuracy of automatic targeting is not enough. In this paper, I calibrate kinematic parameters to improve accuracy of automatic targeting method. The accuracy after calibration is verified by experiment.

In Recent years, CT-guided Interventional Radiology (IVR), which is a mininmal invasive medical procedure, attracts attention. A surgeon observes the perspective CT image and inserts a needle to patient. On the other hand, IVR has a problem that surgeons are exposed to radiation since they operate it close to a CT device. Therefore, we are developing a robot that performs needle inserting by remote control. Current needle holder unit for a CT-guided needle insertion robot has a risk that it might hurt a patient since a needle is always held firmly. To solve this problem, we propose a stiffness changeable gripper by jamming transition. In this paper, we describe a design and prototype of a stiffness changeable gripper to reduce the risk to patients.

By integrating locally operated small surgical robots in a sterilized area, a surgeon can perform safe and accurate robotically assisted laparoscopic surgery. At present, there is no locally operated forceps manipulator that can operate within a small space while providing a wide working area on the abdominal wall. In the present study, a new manipulator united with pivot restraint of gimbal mechanism for the pitch and yaw axes, and a linear slide mechanism for the insertion axis though into the forceps that can act as a third arm for the surgeon was designed. The dimensions of the manipulator are 275 mm * 150 mm * 185 mm. The operating range is 0° to 90° for the pitch axis, ± 45° for the yaw axis and 200 mm for the insertion axis.

We developed a mobile robot with a high expansion camera arm using convex metric tapes. A webcam installed to the top of the robot recognizes human faces and the camera reaches the height of the faces using convex expansion mechanism. The robot detects the distance between the robot and the human using laser rangefinder to approach the human with a mobile robot. The robot can be controlled by human voice, so you’re able to control it from a faraway place. Those systems consist of RT Middleware which distributes architecture helps developers to reuse the robot elements and boosts the reliability of the robotic system.

In this study, an electronic baton system for visually impaired person. The system uses a haptic interface device as a visual alternative interface. In Japan, intermediate blind person due to diabetic retinopathy is expected to increase. Therefore, we propose promoting social participation of the visually impaired person by music support. This paper describe that we Experimented characteristic evaluation experiment of frequency discrimination to clarify the threshold that can discriminate between two frequencies in human tactile sense. As for the result, it was revealed that the frequency at which 75Hz, 100Hz, and 125Hz have a correct answer rate of 80% or more when 125 Hz is set as the reference vibration.

In the task of cocktail party of RoboCup @ Home, when looking for an orderer remembered at the time of ordering and delivering the item, it is necessary to identify the orderer even from distance. This paper proposes the system that can authenticate faces in both short and long distances. The system can get a trimmed face image from the position of the face of the target person using Kinect. By using not only a query image taken at a short distance but also a Gaussian blurred image, it was possible to obtain an authentication rate of 0.5 or more for a distance up to 360 cm.