The types of daily life support tasks that robots can perform are increasing. For a robot to perform spontaneous life-support actions, it must be able to determine what situations exist in its moving environment and decide on tasks accordingly. In this paper, in addition to the situation classification method using a large-scale visual-language model proposed in a previous study, we describe a system in which a person instructs a robot to perform a task according to the situation using a chat interface that can send and receive images and text, and the robot then automatically executes the task. Experimental results show that the system enables the robot to judge the situation and execute the task automatically.

On-site teaching systems that can teach robots to recognize objects in environments such as factories or convenience stores where objects change are important. In this paper, we propose an on-site teaching system for robot recognition using automatic synthesized image generation with object image background removal. The proposed on-site teaching system generates automatic synthesized images using a background removal method to provide data for robot object recognition. Additionally, deep learning is used to construct a model for recognizing objects from the automatic synthesized images. The proposed system shows that the accuracy of object detection using images with background removal is higher than that without background removal. Furthermore, the proposed system was used to perform picking operations on factory shelves using real robots in a short period.

In cooperative driving, in which the driver takes the initiative in driving and the system provides constant assistance, it has been reported that as the strength of the system's assistance increases, the sense of agency decreases. Using a driving simulator, we verified the effectiveness of the proposed auditory feedback during cooperative driving by using a task of deceleration when following a slow-moving vehicle in front. As a result, it was confirmed that auditory feedback tended to maintain or improve the median evaluation of sense of agency. In addition, the auditory feedback tended to reduce the variance, especially under conditions where the intervention rate was low.

Ultrasound diagnosis has been widely employed among diagnostic imaging methods because of its advantages: compactness, low cost, and minimally invasive. However, ultrasound diagnosis places a heavy burden on doctors and ultrasonographers because they need to image the same tomogram continuously for a long time. The examiner’s muscles and skeletons are damaged due to such workload. These symptoms, called musculoskeletal disorders, have become a severe problem. Against this background, the construction of a system to support ultrasound diagnosis is expected. Previous studies have analyzed myoelectric potentials and electroencephalograms (EEG) during ultrasound diagnosis. In this paper, we investigated the relationship between myoelectric potentials and EEG for a cooperation diagnosis support system based on biological signals analysis. As a result, a correlation between an increase in %MVC and an increase in EEG spectral average was observed. This result showed the feasibility of a simultaneous evaluation system of mental and physical fatigue by measuring EEG.

In recent years, research on augmented bodies that extend the capabilities of the human body has attracted much attention. Multi-presence, the topic of this paper, is one of the extended body studies, and is a system in which multiple extended bodies are operated by a single person. Until now, the role of presence in the parallel and multiple manipulation of extended bodies has not been clarified. From the validation test, subjective images with a high sense of presence were more likely to sustain residual attention associated with attention switching than overhead images with a low sense of presence, and task performance was improved. Thus, the results suggest the usefulness of increasing the sense of presence for each task when designing multi-presence.

When a human worker lifts a long bar in collaboration with a robot, the necessary functions required for the robot to accomplish the collaboration with the worker is studied. By observing the lifting work performed by two human workers, a leader and a follower, the essential information for the follower to perform the collaboration with the leader is analyzed. From the analyses of the measured bar trajectories when the visual, auditory, and haptic information are given and not given to the follower, it became clear that the leader's call to the follower greatly contributed to work performance. In addition, it was made clear that the timing of the call at the beginning of lifting are very important for the smooth lifting.

Sense of Agency(SoA) is an indication of human initiative during active movement, and its determination and control is expected to be used for estimating the state and constructing support systems. In this paper, we quantified the dynamically changing SoA continuously using a neural network, and verify the validity of the method. First, an operation delay was set as a condition for changing SoA. Next, we verified the effectiveness of the method by measuring the subjective evaluation values and t proposed method for SoA at valid delay times. As a result, SoA significantly decreased at delay times of 600[ms] and 800[ms]. Then, the proposed method was used for SoA at 800[ms], and no valid results were obtained. However, it was suggested that SoA changes dynamically during machine operation. In the future, the intervention method for delay conditions and the improvement of the proposed method will be the issues to be addressed.

Each type of edema can be a factor in the manifestations of various diseases. However, it is difficult to determine the edema condition because information on the patient, such as the site of occurrence, nature, and medical history, is required. Therefore, this study aims to construct a system that can estimate the type of edema and improve edema according to its characteristics. In this paper, we propose estimating system for edema, which is the core of this research. Using the proposed method, we compared the severity of pitting edema with the proposed model and confirmed the correlation with the spring model.

The number of upper limb amputees in today's society is on the rise due to the increased incidence of diseases caused by longer life expectancy and the increased survival rate from work-related injuries caused by improved medical technology. The number of prosthetic limb amputees is also increasing. Our research is aimed at improving motion estimation and simplifying the associated training to use a motorized prosthetic hand. In this study, we developed a system to identify hand motions using machine learning and verified the number of data required for calibration.

Elderly people account for nearly 70% of all electric wheelchair users, and this number is expected to increase in the future. In addition, the cognitive response speed of the elderly tends to be slower than that of younger people. We believe that increasing cognitive reaction time is a stepping stone to solving this problem. One of the methods is to predict the distance of paw movement before the start of a movement using electromyograms, which are one of the biological signals emitted before the start of a movement. In this paper, we demonstrate the possibility of using machine learning to identify the distance traveled by a fingertip using features based on myopotentials.

We have been investigating a method to operate an electric prosthesis by measuring swelling or hardness changes of muscle with a soft-body pressure sensor instead of myoelectric potential. We found that this method is much more stable and easier to operate than myoelectric potential. In addition, it has been shown that it is possible to grasp silken tofu (very soft and fragile food) with an electric prosthesis by attaching a soft pressure sensor to the fingertip of the prosthesis. We also developed a sensor with three photo-reflectors embedded in a hemispherical balloon-shaped silicone rubber to measure not only the pressure but also the direction of the pressure. In this report, we installed these three elements type soft pressure sensors to the fingertips of a prosthetic hand. We have shown that it is possible to grasp and hold silken tofu more stably using the information of the pressure direction.

This study aims to alleviate the increasing burden on therapists caused by the aging society through the use of robotic rehabilitation. Motivation is an essential aspect of rehabilitation using such robots, yet few studies have investigated it. In this study, we propose a method for evaluating motivation during rehabilitation using electroencephalograms (EEGs), specifically by examining the power spectrum ratio of β/α and event-related potentials of P300. Our results by healthy students indicate that the β/α scores improved in almost all subjects during training compared to rest, and we successfully obtained P300 from EEG data. Thus, our proposed method may provide an effective means of evaluating motivation during rehabilitation using rehabilitation robots.

The develop of low back pain in nursing care work is a serious problem, and various research institutes and companies have manufactured assistive robots to prevent low back pain, but practical issues remain in areas other than assistive performance, such as detachability, miniaturization, and weight. In this study, a basic configuration for generating an assist force to solve the problem with a suspension-type assist robot is realized and experimentally investigated. This paper describes in detail the mechanism design, system design, and control design as the basic configuration, and verifies their validity through experiments.

To reduce the burden of caregivers, we have developed a moving handrail-type device that supports the standing-up motion of a person based on the Assist-As-Needed concept. The trajectory of the moving handrail was optimized to minimize the power of the user’s whole body based on a human dynamical model, using the motion of a person who stands up naturally as a reference. In the Assist-As-Needed method of control, the speed at which the handrail moves along the trajectory is changed according to the force acting on the handrail and the amount of assistance is adjusted so that the handrail provides only the necessary amount of assistance. For this purpose, the control is carried out using a orbit attractor so that the motion is realized without time constraint. Finally, the designed optimal trajectories are implemented in an experimental system with a moving handrail consisting of two-link manipulators and a humanoid robot consisting of four-link manipulators to verify the effectiveness of the control method.

The aim of this study is to develop the wearable walking assist robot that can be used in daily life. The challenge in daily use is stability, which decrease when encountering disturbances while walking. Therefore, it is necessary to analyze the effect of the assist on the user’s stability when facing disturbances and quantify the relationship between them. This study focuses on unplanned gait termination and investigates the effect of assist on user’s physical parameters. The experiment involved comparing parameters obtained from conditions where assist was continued after gait terminate stimulus and where assist was stopped thereafter. Results of experiment suggest that assist that is less than half of the torque required by a user does not significantly change the user’s movement during unplanned gait termination. Thus, it can be inferred that the effect of assist on the stability is considered to be small, at least in healthy individuals.

Unilateral spatial neglect (USN) occurs as a sequela of stroke. Especially, Object centered neglect (OCN) ignores the half side of the object bordering the midline of the object. Conventional assessment methods are considered to be cognitively burdensome to patients due to the large number of attention-obstructing objects. Therefore, we aimed to evaluate the severity of OCN spatially according to the environmental complexity. Symptoms of OCN are characterized by ignoring the missing range of the missing sphere. Using this feature, we developed a system that can present a missing sphere in a VR-Space and evaluate symptoms by location according to whether or not the missing sphere is recognized. Furthermore, this system can adjust the environmental complexity and evaluate the severity of symptoms at each location. Using this system, we tested one patient with USN by changing the environmental complexity. The results suggest that OCN becomes more severe with increasing environmental complexity.

In this study, we develop a walking assistance arm to be mounted on a wheel-type walking training device for the purpose of improving the independent walking ability of elderly people with weakened lower-limb muscles and those with lower-limb disabilities. In this paper, we measured the pelvic assistance force of a physiotherapist for a simulated patient assuming walking with a walker, and analyzed the type and pattern of assistance force during the walking phase in order to know the assistance force required for the assistance arm and its timing. report about

Spasticity, one of the sequelae of stroke, inhibits rehabilitation. We have developed a device called "PDFin" to simulate manual therapy to reduce spasticity in hemiplegic stroke patients and have achieved a certain effect. In our previous work, we have developed “PDFin2” with increased torque and rotation speed by replacing the motor and have verified its safety for healthy people. In this study, we applied PDFin2 to three stroke hemiplegic survivors to verify the immediate effect of MAS improvement and hand extension before and after use. The comparison results with manual therapy showed roughly equivalent effects in MAS and a larger amount of improvement in A-ROM in PDFin2.

We are developing a measurement device that calculates the range of motion of finger joints from 3D point cloud data acquired by an RGB-D camera. The objective of this study is to enable measurement of the MP joints even when there is a misalignment in the yaw angle of the camera relative to the hand. To this end, we developed a function that automatically identifies the location of the hand surface to be measured using Hand Keypoints and acquires the point cloud data. As a result, the required joint angle measurement accuracy of 2 degrees was able to be satisfied if the yaw angle of the camera relative to the fingers is within 20 degrees.

In this paper, the authors propose a method for rehabilitation support in accordance with the patient’s motor intention based on the control of the robot assist and functional electrical stimulation (FES). Motion estimation is performed with a probabilistic neural network by input feature patterns extracted from electromyograms (EMGs). The subject’s motions are controlled and assisted by the developed assist robot and FES control. The experimental results showed that the effective training can be realized by the proposed system, and the system has the potential to provide effective rehabilitation.

A myoelectric hand is a prosthetic hand that reproduces both the appearance and function of the hand. However, due to various factors, the myoelectric prosthetic hand is currently perceived as different from the actual hand, and it cannot be said to be embodied. As with rubber hand illusion, there is a possibility that myoelectric prosthetic hands can be embodied by adding various sensory feedback, but there are many unknowns in the field of myoelectric prosthetics. In this study, I investigated the effects of visual, tactile, and deep sensory feedback on the embodiment of myoelectric prosthetic hands by means of a questionnaire. The results showed that visual feedback had the strongest influence on embodiment compared to the others, with smaller movement delays resulting in stronger embodiment. Tactile feedback promoted embodiment. Deep sensory feedback reduced the sense of ownership but increased the sense of agency.

In this study, in order to provide standing support that considers muscle strength that varies depending on the individual, the support movement must be changed according to the individual's muscle strength status. Therefore, it is necessary to measure individual muscle strength, and many previous studies used electromyographs to measure lower limb muscle strength. However, because the person using the electromyograph must have expertise in measuring muscle strength, ordinary caregivers cannot easily use it in their nursing homes. Furthermore, measurement is difficult because many electrodes must be attached to the subject's body when using the electromyograph. Therefore, this paper proposes a simple method for estimating lower limb muscle strength based on function-specific effective muscle theory. Using the proposed method, a person without expertise can estimate the muscle strength of 3 versus 6 muscles of the lower limb.

In order to assist the elderly to walk independently for rehabilitation of cerebral stroke, a novel rollator equipped with a motor-driven chest pad was developed in the previous study. In this paper, configuration of a measurement and control system for the novel rollator was conducted. Walk measurements for six subjects were conducted using the developed rollator. Time series results and power spectrum were analyzed, and the effectiveness of the motor-driven chest support pad on movement of the lower limb during walking was discussed.

In recent years, the solitary death of elderly people living alone has become a social problem, and demand for a monitoring system to detect abnormalities such as falls and impaired consciousness in the elderly has been increasing. We propose detection system of falls based on differences in the frequency characteristics of floor vibration sounds .We also devised a method to determine the threshold value of the alarm by measuring the staying at a location in the room by estimating the heel contact position. Experimental results showed that the system was able to detect falls with high accuracy. In addition, we were able to obtain effective threshold values for emergency notification for each location in the room.

We intend to realize a new concept of social participation support in which persons with severe motor dysfunctions can operate robots remotely to support caregivers and nursing home users. In this study, we investigated the issues of task efficiency and load caused by operation images. To simplify the recognition of the correspondence between many operation images, we used two methods that merge the supporting image with the primary one. Two merged operation images were compared using a concept-based delivery task which includes movement and alignment, revealing the characteristics of each operation image. In the future, we plan to conduct a detailed analysis, measure additional experimental data on persons with motor dysfunctions, and improve rotational operation.

If elderly people fall, the fear of falling may lead to decreased activity and abnormal gait, which may increase the risk of falling over the long term. Falls often occur during the terminal swing when the heel just before contact with the ground. In this study, we develop a detection algorithm for the terminal swing that does not depend on walking speed using 6-axis sensor. We propose the terminal swing detection algorithm that focuses on angular velocity in the direction of vertical and acceleration in the direction of travel. We used motion capture to verify that the onset of the terminal swing can be detected when the tibia becomes vertical. As a result, it was possible to detect a mean absolute error of 66.1ms or less at walking speeds of 2km/h to 5km/h.

As the population ages, the number of the elderly with systemic diseases like stroke increases. In order to avoid strokes, various wearable walking assistance robots have been developed. However, the most of them are bulky, heavy, and poorly practical. For the elderly users in this study, we developed a compact ankle assistive robot to promote walking exercise. The robot can help the user establish a healthy gait by guiding proper motion and timing while just supporting the ankle joint. To assist users, the robot offers faster-than-ankle movements. With a graphical user interface (GUI), users can change the robot's settings to suit their needs. Furthermore, we proposed a gait-adaptive method for ankle assistive robots to adapt to the user’s changing gait for providing more accurate walking assistance. Finally, the evaluation experiment demonstrated the positive feasibility of gait adaptation. The advantages of the proposed methods include low cost and simplicity of implementation.

There is a rail-type walking support robot, called Yorisoi Robot which is sold by Sanyo Homes. This robot prevents users with walking disabilities from being injured by falling while walking and can thereby support them to become more active. However, users need to operate a path of switch rails with a remote to change their direction of movement, so that they cannot only concentrate on walking though they are disabled to walk. Therefore, we propose a method to estimate a user’s desired direction and automate switching the path of rails by using Long Short Term Memory. An experimental result showed that subjects could successfully walk through switch rails in about 80% with the method.

The aim of this study is to conduct gait analyses on ankles and knee joints on cross-slopes to design walking aids for uneven terrains. To evaluate the motions of lower limbs, we conducted slope-transverse gait (cross-slope walk) experiments on seven healthy subjects (22-44 years old) at four slope angles (0°, 10°, 20° and 30°). Tilted wooden boards were prepared in a measurement room and whole-body measurements were conducted using a VICON optical motion capture system. The experimental results show that there are some significant differences between the motions of the knees and ankles in the upper-/lower-sides.

In neuroscience research using mice, brain activity is observed using techniques such as calcium imaging and linked to mouse behavior. However, because brain activity can only be inferred from the actual observed behavior, if there is a behavior that did not manifest itself, it is impossible to know why. Therefore, if a behavior prediction model is used to predict mouse behavior, and if there is a brain region that is activated only when the prediction is off, it provides strong evidence that there was behavior that was not manifested. In this study, as a preliminary step of the behavior prediction model, a trajectory prediction model was created to predict the walking trajectory of mice. The prediction model was then compared between drug-naïve mice and mice treated with an anxiolytic drug (diazepam), and significant differences were confirmed. This indicates that diazepam administration alters the gait trajectory of mice.

Myoelectric prosthetic hand is a kind of electric prosthesis hand which is an alternative to the lost hand of the upper limb amputee. By using surface myoelectric potential, which is measured from the residual limb, it is possible to perform voluntary operation. However, compared to healthy hands, conventional prosthetic hands have problems such as delay in motion, need for excessive operating force, and sequential motion of prosthetic hands. Therefore, this paper proposed a finger motion estimation method using transient signals of surface myoelectric potential during pre-shaping of finger motion. We implemented a method to detect myoelectric peaks during pre-shaping. As a result of actual experiments, the usefulness of myoelectric information near the peak was verified. Then, we implemented a grasping posture identification method using surface myoelectric potential at the time of myoelectric peak occurrence by pre-shaping.

Conventional methods for motion capture require wearing special suits with markers attached to the body surface, and multiple cameras to capture images simultaneously. In this study, the Microsoft Azure Kinect, a ToF-based depth camera developed for video games, was used for motion capture. Kinect makes it possible to measure the distance from camera to object surface in detail with an accuracy of less than 1 mm. Using this depth data obtained by Kinect, motion capture can be performed without markers. Actually, scenes of hurdle jumping outdoors were captured, and the differences in form between advanced and beginner athletes were analyzed and discussed.

Conveying an individual’s personality is difficult in online conversations, though opportunities for online communication will increase more. Adding information through the processing of facial expressions will become a solution for that. The aim of this study was to obtain statistical models representing associations between facial expressions and personality impressions, which is essential knowledge for processing facial expressions. We collected time-series data of facial movements from 9 professional actors performing self-introduction and processed the data according to Facial Action Coding System (FACS). We also collected personality impressions from 71 participants who watched movies of the actor’s self-introduction. We performed multiple regression analyses with a stepwise method. As a result, we obtained reasonable models for two of the personality impressions: extraversion and agreeableness. The results suggest that extraversion and agreeableness can be controlled by the visual processing of facial expressions.

Human relationship recognition is an important research branch of human behavior analysis, which focuses on interaction patterns and the social relations among two or more persons. We hypothesize that human interactions, the frequency and length of mutual gazes can be used to infer human relationships through video analysis, particularly videos taken by security cameras. With the advancement of deep learning, there has been significant research progress in human interaction recognition for video surveillance systems. However, mutual gaze detection in surveillance camera’s views, in which faces cannot always be seen, still remains a challenge. To verify a part of our hypothesis, we conduct experiments to detect human mutual gazes in videos taken by security cameras and propose a system for mutual gaze detection built on top of a gaze estimator. Two deep learning-based approaches and one angle-based method are evaluated on the collected dataset to compare the effectiveness of the methods.

Because an improper usage of crutches can cause a secondary accident like a fall, it is desirable to train an inexperienced crutch user safely at an early stage of a rehabilitation. In this study, the authors proposed a novel crutch walk training system where an inexperienced crutch user wears a head-mounted display (HMD). Virtual obstacles were presented on the HMD, and the trainee were expected to acquire a skill to cope with them through trainings. Three kinds of experiments were conducted for nine participants, and their results showed that (1) wearing a HMD did not disturb the walk motions significantly, (2) most participants acquired a skill to stop the walk motions shortly, and (3) skills of obstacle avoidance of the participants depended on the direction of an obstacle appearance.

The purpose of this study is to improve the convenience of teleoperation of a small 4-wheeled robot. To this end, we employ a teleoperation technique using a handle-type controller and aim to confirm which angle improves the convenience of teleoperation when presenting a new image to the operator. After creating a virtual space and displaying images from an omnidirectional camera on a Virtual Reality Head Mounted Display, the small 4-wheeled robot is teleoperated using a handle-type controller. In the experiment, a virtual mirror was placed at five different angles on the virtual space calculated from the human viewing angle and the range of motion of the eyes, and compared the results.

This paper describes a verification experiment on the communication coverage area of the low-power, wide-area network LoRaWAN in Ota City, Gunma Prefecture, Japan. The relationship between the topography and the communication coverage area was verified. As a result of the measurement, the relationship between RSSI and distance, and between SNR and distance were clarified. The case studies of LoRaWAN applications are introduced: the measurement of temperature and humidity, and the measurement of the amount of garbage.

The research on force sensing communication technology using haptic devices over the Internet has been actively carried out. The use of this force information for communication between remote locations has been considered as a possibility for communication with a new sense of presence. In this study, we constructed a bilateral control system that enables people to share the same sense of force between remote locations. A force-sharing system was constructed using contact force, the copy robot's own position, and the master robot's position at each remote location by utilizing visual position feedback from a camera.

In recent years, due to the influence of COVID-19 and locomotive syndrome, our opportunities to go out have decreased. Also, due to this, lack of exercise has become a social problem in Japan. We constructed a system which enables us to go out and walk from indoors by teleoperating a wheeled mobile robot as an avatar robot with safe locomotion interface called saddle type device. In addition, we evaluated the system from the point of view of usability. The results of the evaluation showed possibility of improvement of operability and mitigation of mental load by conducting operation practice in advance.

In order to collect information in a disaster-stricken closed space, it is necessary to be able to remotely control a mobile robot without losing its position in the event of environmental changes such as collapsed objects. Indoor positioning methods include odometry and SLAM. Odometry position estimation has the problem that errors accumulate due to the movement of the mobile robot over obstacles scattered after the disaster. SLAM also requires more information to be collected by sensors when obstacles are scattered, and more computation time is needed for object detection. As a result, it is difficult for SLAM to cope with environmental changes. In the closed space where conventional position estimation is difficult as described above, we propose a position estimation method for mobile robots using Ultra Wide Band.

In closed spaces after a disaster, information gathering using tele-operation of mobile robots is widely studied to minimize damage. One tele-operation method called "Robot Wireless Sensor Networks" that uses wireless LAN and relays has been considered for tele-operation of mobile robots. However, a challenge arises from the risk of mobile robot distress caused by wireless LAN disconnection. To address this, we have discussed a tele-operation method using multi relay with LoRaWAN to prevent mobile robot distress. In this paper, we experimentally confirmed that tele-operation with multi relay using LoRaWAN for command-type tele-operation is possible over long distances of up to 2.0 km, where wireless LAN communication is difficult.

This paper describes study the Robot Wireless Sensor Networks (RWSN) using mesh networks. RWSN is a disaster information gathering system using mobile robot and sensor nodes which relay networks. The robot can be tele-operated from a safe location and gathered information without human entry into the closed space. In RWSN, it is necessary to construct a temporary communication infrastructure using WSN that there is possibility that the existing communications infrastructure is unusable due to the disaster damage. Therefore, we examined the feasibility of operating with dynamic routing by incorporating a mesh network into RWSN.

It is known that human pupil is related to own affects or interests and becomes more attractive when it is expanded. Focusing on effects of the dilated pupil in human communication and interaction, we analyzed the relationship between the speaker’s speech and his/her own pupil response and confirmed that the pupil dilates in synchronization with the speech. Based on the analysis, we developed the pupil response interface that expresses various pupil responses such as dilated or contracted pupil by speech input. In addition, we developed the speech-driven embodied listening system by mirroring of pupil response in synchronization with the speaker’s speech. This system expresses a positive listening attitude by the pupil response and the nodding response. In this study, we evaluated impressions that the system mirroring the pupillary response acted as a listener. The effectiveness of the system by mirroring the pupil response was demonstrated by sensory evaluations.

It is well known that the tails of dogs function to communicate with their pack mates or owners and to express their emotions by changing the tail positions and the way of wagging their tails. For this reason, there is a growing trend to use tail wagging as an objective indicator of dogs’ emotions in experiments in the field of biology. However, since the breeds and personalities of dogs participating in experiments generally vary, each dog’s tail wagging often differs for the same emotion. Therefore, this study aims to develop a personalized emotion classification system that extracts the wave characteristics of tail wagging acquired from pose-estimation software, SLEAP, and assigns emotion labels based on the situations.

This study proposes a method to quantify the educational effects of "using robot teaching materials" from the student's classroom behaviors. In this report, we developed a prototype device that simultaneously measures actions and speech using SPRESENSE and measured the actions and speech of fifth-grade students during an 80-minute robot programming class. From the discussion, we showed that there is a relationship between the similarity of actions and the degree of cooperation within a group.

In this paper, we describe an identification of backpack-type touch devices for tag-play with a robot. When a robot is about to touch a player in the tag-play, the device carried on the back of that player moves an arm to touch the player. If there are multiple players with the devices, the robot cannot identify which device the target player is carrying. Therefore, we propose a method so that the robot identify the device of each player by acquiring the distance between the robot and the device using RSSI. We conducted an experiment, and verified that the robot could touch correct player when the robot played tag with multiple players.

The bi-articular muscle-tendon complex in animals contributes to the dynamic motion such as high jumping or fast running. Our research group has developed legged robots with a mechanism that mimics the bi-articular muscle-tendon complexes on animal’s limb. However, its effectiveness of the mechanisms in forward jumping has not been verified yet. In this paper, forward jumping trajectories for a legged robot with bi-articular muscle-tendon complex mechanisms were generated. In the forward jumping simulation, the legged robot with the mechanisms jumped longer distance than those without the mechanisms. Based on the generated forward jumping trajectory, experiments using the legged robot with the bio-inspired mechanism were also conducted. Forward jumping utilizing the effect of the bi-articular muscle-tendon complex mechanisms was realized in the experiments.

This paper presents a method for locomotion generation of a humanoid robot walk by reinforcement learning (RL). Specifically, the RL is designed to learn a policy for outputting the joint torques that perform periodic combined motions of limbs, aiming at acquiring a whole-body locomotion. The effectiveness of the learning method is validated on the MuJoCo simulator using the small humanoid robot ROBOTIS-OP3. We report the results obtained by adjusting the reward terms accounting for performance for biped and quadruped locomotions and other parameters.

In this study, we attempt to acquire robot behaviors that will improve a familiarity with humans. Based on the results of that the interaction affects human familiarity with robots, we propose to evaluate the robot behaviors by using Deep Q-network. In addition, we build a type of non-living robot to eliminate the influence of appearance.

We propose to apply machine learning to acquire the optimal flapping behavior for Micro Air Vehicle (MAV). In this paper, we create a model of dragonfly in a virtual physical environment and build a controller for the flapping motion using an artificial neural network. Through several simulation experiments, it is possible to verified that our proposed method acquires a controller for the target-tracking behavior.