Recently, welfare vehicles are widely used among the aged. However, it is hard for inexperienced beginners or the aged to control the welfare vehicle in narrow places such as elevators and corridors. For these problem, the cognitive and driving ability is enhanced from the simulated objective view point with a Head Mounted Display (HMD). And the alternation of driving assistance and the automatic driving system is required. In this paper, we focus on an automatic driving system using wearable MR-device which has collision avoidance function using spatial mapping. The driver rides on the vehicle and controls it using the virtual preceding vehicle with HMD. The obstacles are mapped real-time with the spatial mapping function installed in HoloLens, and the virtual vehicles of the steering assist system are driven automatically using A* (A-Star) using NavMesh system. The effectiveness of proposed avoidance system is evaluated in a narrow corridor.

We propose an absolute velocity estimation method of surrounding objects using stereo high-speed vision. Surrounding objects are detected and tracked at high-speed by stereo inverse perspective mapping (stereo IPM), meanwhile relative positions are estimated. Relative positions at two adjacent frames are transformed to the same coordinates by using ego-motion information to compute absolute velocity. Ego-motion is estimated by computing relative motion against the ground surface using vacant area detected by stereo IPM to achieve robust and accurate estimation. We demonstrated the proposed method greatly improved the accuracy of absolute velocity estimation through experiments in a vehicle scenario.

In recent years, researches on object tracking, which targets children’s monitoring and automotive applications, have been actively conducted. In object tracking, modeling the motion of an object to be tracked is an important task for determining the performance of the tracking algorithm. Therefore, a method was developed to model the motion of the tracked object using optical flow and track it with a particle filter. In this paper, the comparative experiments of tracking accuracy with the conventional technique are conducted by switching the model by an optical flow, and the experimental result is reported.

When building autonomous mobile robots, the challenge has often been to create a system capable of functioning as intended to even if some components fail. This research aims to approach the problem by fusing multiple sensor data using an extended Kalman filter and probabilistic occupancy mapping. A nonholonomic feedback controller controls a mobile robot which uses the pose obtained from the extended Kalman filter as a reference frame for the sensors. We use ROS framework as a base to implement this on our system.

Over the past decades, 3D jobs (dangerous, dirty, demanding) are expected to be replaced by autonomous robots. At a theme park, surveillance, cleaning, and guiding tasks can be regarded as 3D jobs. This paper aims to develop an autonomous tour guide robot system at a large theme park. One of the characteristics of the tour guide robot we developed is the use of QZSS (quasi-zenith satellite system) and CLAS (centimeter level augmentation service), which realizes the centimeter-level positioning for autonomous service robots. We also implemented autonomous tour-guide system including path planning, collision avoidance, and voice explanation. Experimental results at the theme park showed that QZSS has sufficient localization accuracy for the tour-guide task, and the tour-guide experiment was successfully carried out.

In this study, we propose a method of pipeline robot’s navigation based on fuzzy logic control (FLC) for passing through elbows or T-junctions. A CCD camera is used for locating the region of interest (ROI) in elbow or T-junction. Moreover, ROIs formed by reflection of LED light and the edge of pipe’s dark hole are considered as input variables in the FLC system. By analyzing outputs of the system, we can control the robot’s speed and yaw angle in real time. Compared with conventional studies on pipeline robot’s navigation method, the proposed method can be more precise and faster by using FLC algorithm and analyzing ROI with fewer sensors. Finally, we performed a simulation validation, and the results showed that the robot could adapt to known pipe environments and realize navigation in straight part, elbow, and junction of pipe.

There should have some mirrors in the indoor environments. Autonomous mobile robots should detect mirrors not as wall, but as mirror, because there are more advantages that can use virtual images from mirrors. This should be able to use in unknown environments for robots. In this paper, we suggest algorithm of flat mirrors detection system using RGB-D data and develop preprocessing part of whole programs, and experiment of calculating the coordinate and orientation of one of the mirror surface point. Our goal is to get coordinate and orientation of whole of a mirror and reflect the result on the environmental map for later uses.

We have been developing a guidance robot “EM-Ro” that works in ECO35 muffler museum. EM-Ro has already succeeded in taking visitors in ECO35 muffler museum autonomously and describing display items by voice sound. In this paper, visitor detection and autonomous following algorithm using an environmental map, as an optional function, was developed successfully. This visitors-following algorithm was used only by 2D LIDAR and didn’t need to change drastically an autonomous traveling system that we have developed. The experimental results showed that EM-Ro could follow visitors smoothly, and re-detect visitors even when entering in its blind spot.

In this paper, a position measurement method for miniature mobile robot is proposed. the method is based on position tracking device for VR. This position information is able to apply kinematic analysis of mobile robots. Since the device can measure 3 different positons simultaneously, the posture of mobile robot can be estimated. This position information is compared with the posture which is estimated by inertial measurement sensor.

Avoiding obstacles is one of important function for outdoor mobile robots. Although detecting three-dimensional obstacles and avoiding them have been studied, planar obstacles, such as puddles, have been ignored. Avoiding puddles is important to prevent electric problems, odometry error, and feeling displeasure. However, it is difficult to detect puddles by using only visual information. Because puddles change their shape and their color due to environmental conditions. This paper proposes the method to detect puddles by using laser reflection intensity and path planning method to avoid the puddles. Laser reflection intensity of various road surfaces were investigated, and the effectiveness of the proposed method was confirmed through real-robot experiments.

Safe landing system is critical for an unmanned aerial vehicle (UAV) for transportation, land surveying and exploration. In particular, a safe landing system to unknown and unstable ground is needed in order to expand the flight area. Spatio-temporal resolution of 3D sensing is one of the most important performance measure when UAV recognizes detailed ground shape and the dynamics for safe landing. In this paper, we propose a 3D sensing system that adaptively controls the measurement area and improves the spatio-temporal resolution of 3D sensing. In experimental setup, we evaluate the improvement of the performance by refining measurement area.

Bats are the one of unique sensing animals who perceives the 3-dimensional space by an echolocation. From an engineering point of view, echolocation design seems to be quite simple owe to be consist of only one transmitter and two receivers. Despite of this, bats accomplish not only acrobatic 3-dimensional flight but also flight for capturing prey in a real time. In this study, we constructed a method for 2-Dimensional localization process with binaural frequency beat information inspired by bats. In addition, we also evaluated the proposed method by the practical experiments with automatic sensing system equipped with one transmitter and two receivers.

The positional information can be used for navigation of the robot, and it is very important information. GPS is often used for position detection, but high precision measurement is difficult. And also, GPS cannot be used indoors. Furthermore, for position detection by the camera, since it is necessary to analyze by coordinating a plurality of cameras, much cost is required. In this research, we describe the development of a position estimation system using a small size wireless module. The position is estimated from the radio wave intensity with a plurality of antennas installed at predetermined positions. The radio field intensity is nonlinear with distance and it is difficult to calculate the distance, therefore neural network was used for distance estimation.

We developed a mobile robot simulator based on Unreal Engine 4 to help software development of Tsukuba challenge. The simulator simulates a area nearby Tsukuba city hall, wheeled mobile robot with a velodyne VLP-16 lidar and a inertial sensor. Experiments with our two different navigation programs for Tsukuba challenge 2017 and 2018 are shown. Both systems were successfully localized and controlled the simulated mobile robot.

We developed a high expansion rate arm 2 with convex metric tapes. The robot is installed three convex tapes and motors that rotate and expand the tapes vertically using RT Middle ware. Also, the top of robot is installed an acceleration sensor and it shows the inclination of the top of robot constantly. Here, we newly developed the expansion system to get wide range motion and expansion using tilt angle controlling witch adjusts each length of convex mechanism. We verified the face angles and expansion heights with some loads by basic experiments and compared with the first expansion mechanism.

Distributed tactile sensing with multi-fingered hands can provide high-dimensional information for grasping objects but it is not clear how to optimally process such abundant tactile information. The current paper explores the possibility of using a morphology-specific convolutional neural network (MS-CNN). In particular, uSkin tactile sensors are mounted on an Allegro Hand, which provide 720 force measurements in addition to 16 joint angle measurements. Consecutive layers in the CNN get input from parts of one finger segment, one finger, and the whole hand. Since the sensors give 3D (x, y, z) vector tactile information, inputs with 3 channels (x, y and z) are used in the first layer, derived from the idea of such inputs for RGB images from cameras. Overall, the layers are combined resulting in building a tactile map based on the relative position of the tactile sensors on the hand. 7 combining variations were evaluated, and over 95 % object recognition rate with 20 objects was achieved, even though only one random time instance from a repeated squeezing motion of an object in an unknown pose within the hand was used as input.

In this research, a newly developed slant wheel system as a compact omnidirectional mechanism is mounted on a mouse type self-propelled haptic device. It is confirmed that the characteristics motion with the STR motor plays a role of steering and moving motion, and also that the slant wheel system causes the omnidirectional movement through some experiments. These experimental results are roughly consistent with theoretical values.

Tactile sensation changes with time, there are individual differences in tendency of temporal change of tactile sensation. However, there are no research examples that evaluate and analyze tactile sensations as time series data. In this study, subjective assessment of skin lotion was carried out using the Temporal Check-All-That-Apply (TCATA) method to measure tactile sensation data of time series. We also analyzed individual differences in time series data using cluster analysis. As a result, four groups were classified from 39 participants, and each of the four clusters showed a different tendency of temporal change of tactile sensation.

Force feedback devices are attracting attention as a means to apply virtual reality and augmented reality to sports, entertainment, and rehabilitation. We have developed a force feedback device Using pneumatic artificial muscle and magneto-rheological fluid clutch. However, there were problems such as increase in the weight of the device and complication of control system. In order to solve these problems, a new force feedback method using an elastic spring and a magneto-rheological fluid clutch was proposed. The force feedback characteristics of the force feedback method were modeled. A 1-degree-of-freedom force feedback device using a new force feedback method was developed. Modeled characteristics of force feedback were confirmed by conducting experiment for rendering elasticity using the device.

In this paper, we propose a new object handling device for robot with force detection capability by coating conductive thin film polymer poly (3,4-ethylenedioxythiophene) (PEDOT) [1] on the surface of conventional vacuum suction cup. The applied force can be detected by the resistance change of the added conductive thin film. The feature of our research is that it can work not only as a tool that the grasping tool but also as a force sensor. The PEDOT thin film was prepared by a dip coating process that allows the thin film to be directly coated on a three-dimensional (3D) surface. We fabricated a sensable suction cup．

This paper describes a triangular-shaped soft tri-axis tactile sensor for implementing on the complex curved surface. The sensor is composed of a silicone rubber embedded a ferromagnetic marker and three inductors arranged in a triangular lattice pattern. In this structure, applied tri-axis force can be estimated by measuring the change in inductances caused by three-dimensional displacement of the marker. As a preliminary experiment, we investigated the basic sensor response, i.e. the changes in inductance versus the displacement of the marker. The experimental results indicate that the estimation error was large in a triangular lattice arrangement of three inductors when we employed the sum and difference of the measured inductances to estimate the marker displacement. We also found that the shape of the marker affects the estimation result of the marker displacement, and that the estimation error was reduced by using an inverted triangular shaped marker instead of a circular shaped marker.

Sensorized realistic hands are important for android robots to realize physical human-robot interaction. In this study, we developed a prototype hand with tactile sensors and bone parts for a child android robot Affetto. Soft inductive tactile sensors were mounted on the palm bone part and free joints were implemented on fingers. We investigated the sensitivity of the sensors and found that they can detect slight human touch. Additionally, we interviewed 13 human participants on tactile impression of the hand and found that its realistic impression was not deteriorated by the embedded tactile sensors but deteriorated by surface properties such as tackiness and tension, and by movable directions of finger joints.

In this study, we discuss hardness perception in a relationship between a tapping motion onto a virtual floor and passivity of a virtual coupling (VC). Even in a tapping motion with short contact time, an operator can feel not only strength of hardness but also qualitative difference of hardness. We consider the difference in the quality of hardness from the point of view of passivity analysis of VC using a design method of VC that adjusts desired passivity in various frequency ranges by using multiple fractional derivatives. Then we construct a system to display haptic feeling of VC with passivity having a peek in various frequency range.

As a method of imparting a tactile sense to the VR experience, a method has been proposed in which the experiencer actually touches an object of the same material as the object visually displayed in the VR space. However, with this method, it is necessary to prepare the same number of objects as the type of the object in the VR space.In this research, we aim to measure the boundary at which humans perceive the difference between visual information and tactile information when presenting visual information by video and tactile information by real object at the same time, and we conducted measurement experiments using constant method and limit method went.

Perceptual sensitivity of the skin depends on skin deformation. Conventionally, perceptual sensitivity and skin deformation are measured independently. So, it is difficult to investigate the direct relationship between skin sensitivity and mechanical properties. In this study, we propose a method to realize simultaneous measurement of skin deformation and sensory sensitivity by applying suction pressure on the skin, which is commonly used for determining the mechanical properties of the skin. We introduced an oscillation of 10 Hz to the suction pressure to generate enough stimuli for the sensitivity measurement. This paper reports an overview of the proposed device and a pressure control method. We also conducted a pilot study to assess the skin deformation and perceptual sensitivity on the forearm skin with 13 subjects. Experimental results showed that the amount of skin deformation had a high correlation with the correct answer rate.

We developed a real-time waveform generation model for representing impulsive collisional sensation using the RungeKutta method based on a contact mechanics and vibration propagation model, which were proposed previously by the authors’group. The developed model can generate vibrations depending on the stiffness of hitting objects and hitting speed for VR applications. As an example of applications, we measured vibrations occurred on the drumstick when it struck a snare drum and determined the parameters. Comparisons with the measured data demonstrated that our model could reproduce similar vibration waveforms.

Human touches his/her skin using fingers on a daily basis. It is important to understand the tribological phenomena while fingers slide over soft surfaces such as skin. We investigate the coefficients of friction when a bare finger slides over artificial skins of different softness levels whereas most of the earlier studies employed a rigid friction probe sliding over the skin. We found that the coefficient of friction is constant irrespective to the finger’s normal force when the artificial skins are harder than the finger pad. In contrast, the coefficient of friction decreases as the the finger’s normal force increases for the artificial skins softer than the finger pad. Furthermore, the coefficient of friction itself and its speed dependence component become greater for the softer artificial skin.

Tactile perception of textures have been explained by the properties of either of normal or shear forces induced between the fingertip and material during tactile exploration. However, we have found that the stochastic relationship between these two axial forces differ among textures in our previous studies. The phase difference between the two axial forces is one of the such relationship. To investigate whether the phase difference between the two axial forces affects the tactile perception of textures, in this study, we investigated the discriminabilty of the phase difference. We found that phase difference between two axial forces can be discriminated in roughness textures with large spatial wave lengths. This result can be applied to develop the method to present virtual texture by combining two types of tactile stimuli.

To reproduce realistic tactile sensations on a tactile display, it is important to elucidate the relationship between skin deformation and tactile sensations. For this purpose, we previously proposed a device to measure 3-dimensional skin deformation but it was not evaluated using actual skin. In this study, we improved the device to address the issues in measurements using real skin. Then we conducted an experiment involving actual skin deformation measurement. The overall shape of the result agreed with our expectations but it had large errors and noise. Therefore, skin deformation measurement using this device is possible but it is necessary to reduce errors and noise to use this device for actual measurement.

This paper presents the development of a tactile sensor that has a simple structure and can be sprayed over three-dimensional shapes such as curved surfaces. The sensor can be modeled by the distributed constant RC circuit, and the contact point is detected as a signal phase delay caused by it. It was found that the accuracy of position detection using the time delay between signals decreases depending on the input frequency. And so a new position detection method that does not depend on frequency was proposed. The method is based on finding the input-output relationship between the contact point and the output terminals using the z-transform. A method of reconstructing the voltage waveform generated at different contact points from the corresponding inverse transfer function was developed and from the error between these waveforms the position can be estimated. Discussions regarding details of the system and test results are presented.

It is significant to develop a flexible rubber as haptic sensor because artificial outer skin on a robot is made of rubber. The rubber skin might be ideal if it has hybrid functions. It is feasible to have the rubber many functions involving Piezo- and photovoltaic effects, and capacitance by utilizing a magnetic compound fluid (MCF) into a diene-type rubber, natural rubber (NR) or chloroprene rubber (CR). In addition, at the fabrication of the sensor with using MCF rubber, the adhesion of electrodes or electric wires on the rubber is so important that the conductivity between the rubber and the electrodes should not be reduced. This paper presents not only the new adhesion technique between rubber and metal but the fabrication of the rubber sensor.

This paper describes mechanism design of an in-pipe mobile robot using a wheel drive mechanism. The wheel mechanism, which is driven by planetary gears and worm gears, can both drive the wheels and press them the wheels against the pipe wall using only an actuator and simple mechanisms. In the new mechanism, the design of reducing the total length of the wheel traveling mechanism by changing the arrangement of the motor so that it can run on a curved pipe or branch pipe was made. The prototype robot has diameter of 45[mm], length of 62[mm], and weight of 60[g].

The efficient generation of uniform core-shell droplets have an important role in industrial fields. In this research, mono-disperse core-shell droplets have been generated using a flexural bolt-clamped Langevin-type transducer and two micropore plates. Core-shell droplets were generated in the air when water and silicone oil were used as the core and shell phases. When the applied pressure of core phase, the applied pressure of shell phase and the vibration velocity in micropore were 200 kPa, 150 kPa and 8.2 mm / s, the average diameter and the coefficient of variation of droplets were 207.7 μm and 1.6 %. This result shows that the developed device is effective for the generating of mono-disperse core-shell droplets.

Unmanned aerial vehicles (UAVs) or regularly known as drones, are now a common feature of our modern society. They are highly useful for collecting information at places where people could not reach especially during natural disasters. But while the current generation of UAVs are undoubtedly useful, the need for smaller and more agile systems that can be easily navigated in narrow areas is starting to get more attention all around the world. This need leads to the development of Micro Aerial Vehicles (MAV). Thus, in this paper, we describe the trial production of micro ultrasonic motor and micro propeller for developing the smallest MAV in the world.

Small flying robots are expected to play an active role in various fields such as disaster sites, inspection in plants. Those robots are required to be smaller and to drive in wires operation. In order to realize this, we focused on self-excited electrostatic actuators. Because those actuators can drive in oscillatory motion with a simple structure and DC power supply, it is considered to be suitable for small flying robot which enable wireless flight. We improved output efficiency by giving elastic element and confirmed the flapping motion of a small fling robot using the improved self-excited electrostatic actuator. In this paper, we experimentally verified the lift force using a prototype.

We proposed self-folding method of electronic substrate using hinges with Kirigami structure for stretchable electronic devices. Stretchable electronic devices using folded structure have been researched. Self-folding methods of electronic substrate are needed to miniaturize a folded structure. However self-folding with electrically connected at hinges is difficult to fold in both mountain and valley directions. We realized the self-folding with electrical connection at hinges in both directions of the electronic substrate by using hinges with Kirigami structure. Furthermore, we revealed relationship between hinge structure parameters and curvature, and realized curvature control by design of hinges with Kirigami structure. We realized self-folding at large curvature compare with hinges with simple beam structure, based on the relationship.

We have modeled Kirigami structure in order to theoretically design rigidity and breaking strain as a device by changing parameters of Kirigami structure. Kirigami structure can change the rigidity and breaking strain as a device by change the dimensional parameter of Kirigami structure. However, the relationship between the dimensional parameters of Kirigami structure and mechanical characteristics such as rigidity and breaking strain as a device has only experimentally clarified. Therefore, we clarified the relationship between the dimensional parameters of Kirigami structure and the rigidity and breaking strain as a device by modeling Kirigami structure. We compared the value derived from the model and measured values calculated by the tensile test in order to investigate whether the model can be applied to the Kirigami structure. As the comparison result was in good agreement, the result indicates that the model can be applied to the Kirigami structure.

Communication robots should be better as small as an infant for reducing the user's feeling of pressure. And non-verbal communications (such as gestures) are also important for smooth communication between human and robots. In this paper a small robot hand for gesture communication by using SMA (Shape Memory Alloy) is proposed. Each finger is driven by an individual SMA wire. Required specifications of size, operation angles and response times for the robot finger were decided according to the human finger. These required specifications were fulfilled by considering the geometric designs and heating/cooling power control. The five finger hand is also under fabricating. The five finger hand is now under fabrication, and designing how generate and control its gestures.

Self-propelled microswimmers have significant potential for medical applications. Our concept uses biofuel cell (BFC) and electroosmotic propulsion (EOP). This EOP mechanism suggests that a smaller body would have a larger velocity. For the larger velocity, we need a design and fabrication method suitable for the body size comparable to red blood cell. We used multiple types of polymer composite layers with different conductivities and the two photon polymerization method, and succeeded in fabricating the smaller prototype and identified its conditions.

Learning of the robot movement with reinforcement learning (RL) has attracted attention, and improvements of various RL methods have been carried out intensively. With conventional RL methods, however, a complicated task takes a long learning process, which is problematic in the robotics domain. In this paper, we focused on the compositionality of policies of Soft Q-learning (SoftQL). With SoftQL, it is possible to compose multiple already-learned policies and execute compound tasks efficiently. However, in the SoftQL, the action-sampling procedure and learning algorithm are complex due to the continuous action space. In this paper, we applied the SoftQL to a maze-solving problem which has discrete space and investigated its performance and computational tractability for discrete-space problems.

In this research, we propose a motion generation method for inserting rod-shaped pegs from the neighborhood of the hole by using only the joint angles and current values measured from the robot arm’s motors. Conventionally, expensive torque sensors are used about peg-in-hole. However, we confirmed that the peg inserting motion can be generated by learning a neural network using not torque sensors but the motor angles and current values. The learning model is a recursive neural Network integrated and learned by time series data composed of joint angles and current values made with a remote teaching system and it predicts the next motion at a certain time by its position and variance. We confirmed that flexible peg insertion can be realized from multiple initial positions by feedback control with proposed method using motor's joint angles and current values at each time after learning.

Although value function-based Reinforcement Learning (RL) has been successfully applied for a variety of tasks as well as with policy search, manually designing appropriate reward functions for such complex tasks as robotic cloth manipulation still remains challenging and costly. Inspired by the recent success on Generative Adversarial Imitation Learning (GAIL) in policy search, which allows an agent to learn near-optimal behaviors from expert demonstrations without explicit reward function design, we explore an imitation learning framework for value function-based RL approach. The generator of GAIL needs to have both the smoothness of policy update and the diversity of the learned policy. We first propose a novel value function-based RL method, Entropy-maximizing DPP (EDPP). Then the corresponding imitation learning framework, P-GAIL is developed. In order to investigate the performance, we applied P-GAIL to the flipping a handkerchief task.

Recently, a robot that can adapt to various environments has been expected, and the necessity of controlling complicated systems has been increasing. However, there are some issues for controlling the entire system with one controller, so flexibility could be necessary for the controller. Therefore, we examine the control method from the motion data based on behavioral approach. In this study, we extracted and analyzed the latent features of sensors’ and actuators’ data of the robot, and investigated the role of that each feature values play in robot locomotion. We performed experiments using a developed quadruped robot that can acquire multi-point motion information as the movement data, and extracted and analyzed feature values from motion data using the auto-encoder. As a result, we found that it is possible to realize various movement patterns from one specific motion learnings.

This paper proposes reinforcement learning with hyperbolic discounting. In general, return and its expectation, i.e., value function, are defined as cumulative rewards with exponential discounting due to mathematical simplicity. Animals, however, show behaviors that cannot be explained by the exponential discounting, but can be explained by the hyperbolic discounting. There is therefore no doubt that some profits can be obtained by changing the exponential to hyperbolic discounting. Combining a new temporal difference error with the hyperbolic discounting in recursive manner and reward-punishment framework, which is also biologically plausible, a new scheme to learn the optimal policy is derived. In simulations, it is found that the proposal outperforms the standard reinforcement learning, although the performance depends on the design of reward and punishment. In addition, the averages of discount factors w.r.t reward and punishment are different from each other, like a sign effect in animal behaviors.

Robots working at warehouses or living environments must avoid obstacles including people and robots itself. A lot of conventional researches on obstacle avoidance have problems that paths must be designed by the engineer or that the computational cost of path planning is high. In this research, we propose a novel obstacle avoidance without any prior design or the high computational cost. Based on manifold hypothesis, we implemented and trained Generative Adversarial Networks (GAN) to obtain lower dimensional representations from postures of 2 link arm robot which avoid obstacles. We confirmed the adjacency of the latent representation corresponds to that of joint angle after training of GAN. We also confirmed that when the trajectory on latent representations was drawn, it was mapped on the joint angle space and the smooth trajectory avoiding obstacles was planned.

There are various kinds of information in human-human interaction, but it is difficult to process all of them. This research aims at understanding mechanism of information selection and proposes a computational model based on prediction error minimization (PEM) mechanism. The PEM mechanism is a hypothesis that explains brain function as a minimization process of prediction error. Although it has been demonstrated that action generation by the PEM mechanism is important in interactions, the situation dealing with multiple information has not been considered. This research focused on uncertainty of information and used a recurrent neural network model that learns to predict the next state of information and uncertainty. In the experiment, ball-rolling interaction between a robot with the model and that operated by an experimenter was conducted. The robot switched its action in accordance with only the information with small variance. This result suggests that uncertainty is important for information selection.

It is important for collaborative robots to share goals with a human partner, but considering all the possible goals in advance is difficult. Therefore, collaborative robots are required to generate correct action towards new unlearned goals. In this research, we focus on a method of generating an action by providing a goal image. In order to acquire the relationship among the goal images, the corresponding visual information and the robot’s joint angles, Long Short-Term Memory Recurrent Neural Network was used. We confirmed that a robot with our proposed method was able to generate correct actions even when unlearned goal images were given.

Recently in aged society, the role of oral care is getting larger and larger because keeping oral health can bring aged people higher quality of life. So it is quite important to develop an educational environment in order to bring up nurses have practical oral care skills. In order to serve students in nursing an effective oral care training environment, we have been developing an oral care simulator. This paper describes the second prototype of the oral care simulator and its software. The system consists of 3 components; denture models with pressure sensors printed by 3D printer, a micro computer performs A/D conversion for signals, and a software visualizes recorded data. Many improvements from the 1st prototype are made including increased similarity to real human, and downsized devices. Calibration for the sensors was conducted, and we found more improvements for accurate detection of brush stroke weight are needed for practicality.

Massage is used for various purpose. But recently, the deterioration of the quality of the massager is regarded as a problem. Therefore, quantification of massage technique is important. In this research, the purpose is development of the massage learning support system using a pneumatic soft device with position sensors. We use pneumatic soft device as an arm model. We use pressure sensor and position sensor for sensing of force and position. And we use “OpenGL” for displaying visually easy to understand. The Basic experiment of the massage device shows effective possibility. In this paper, we will describe the system configuration.

In foreign countries, most of the robotic care equipments developed in japan are categorized as medical devices. So, we require attention that we cannot commercialize the robotic care equipments as it is, in some foreign countries. We need to comply with some regulations for medical devices and so on in that country. In this paper, we introduce examples that we complied with some regulations for selling robotic care equipments in Hong Kong and Taiwan.