In this study, we constructed a system that enables human beings to adjust interactively the handing over of the robot for each human posture and conducted an experiment to verify its feasibility. This system focuses on the human's ease of receiving the object, and interactively adjusts the handing position and the hand direction of the robot (handing angle). Fuzzy inference is used to determine these adjustment amounts, and a neural network is used for posture classification. A human applies a force as a fuzzy inference input to the force sensor at the hand of the robot when receiving an object, and the position and angle at the next handing are determined.

In order to operate service robots in a human-robot coexisting environment, it is important that robots need to avoid people carefully and smoothly. In this paper, we propose a noble control method, which can generate appropriate motion for each person’s behavior. The proposed motion control method is based on DWA (Dynamic Window Approach), which is used for obstacle avoidance scheme using the optimization of several objective functions. By changing the parameters of DWA based on SFM (Social Force Model) expressing human intentions, motion control considering each person’s behavior is realized.

In this paper, we realized the transfer robot with wide motion range manipulator by the lifting mechanism. This lifting mechanism has a pantograph jack system and transforms the table height with 200mm to 900mm. We verified the stability of table and the change in range of motion by experiments.

Due to globalization, the number of businesses between foreign countries has increased. Because of this, many companies are promoting telework as a way to work in different locations around the world. As a part of telework, telepresence robots are expected to be used more widely. However, commercial telepresence robots now available, have no arms or components to address an affective gesture. Thus, the only effective way to address the remote operator’s intention is to use the voice, which might lead to interrupt the flow of discussion. In this research, we aim to improve telepresence robot’s presence and smoothly gain a chance to speak during a discussion. For this we developed a system, where a RGB camera on iPad detects the remote operator’s posture and send signal to telepresence robot to conduct intentional motion during a discussion. As a result, from PAD and survey we were able to successfully fulfill our aims.

This paper describes a method of estimating human pose when it is fully or partially covered by cloth objects like blankets. Human pose in-bed estimation is challenging due to its unpredictability of the covered parts. Therefore, we used a 3D computer graphics application Maya to create datasets. They are created by changing the human pose, the thickness of a cloth object, and the location and the angles of a camera. We simulate the situation where a fixed camera on ceiling or a mobile robot with a depth camera estimates human poses under a blanket. Using the datasets and U-Net, a network for semantic segmentation, we confirmed that 94.02 % of pixels of the test data (overhead viewpoint) and 88.16 % of pixels of the test data (diagonal viewpoint) were estimated correctly. However, the results of the estimation in real world were not good enough to recognize the human pose.

This research aims to construct a semi-automatic remote control system, which utilizes the autonomous function of a robot in a user’s remote control. One of the challenges in remote operation is the latency between the user’s operation and the robot’s action. This problem is solved by using information about hand motion and the functional meaning of the object (affordance) for estimating the user’s intention early. In the previous research, an early estimation of offline operation intention has already been performed. In this research, we move an actual robot by estimating the user’s operation intention from online inputs. We then evaluate the operation speed and the accuracy of the robot and confirm its usefulness. We obtained the average accuracy of 70 [%] and operation times of 3.954, 4.169, and 4.439 [s] for each of 4, 7 and 10 frame lengths.

This paper describes a system method that generates optimal questions by mobile service robots for disambiguating user instructions in Bring-me tasks. The task of the robot is to bring a specific object at a remote location specified by the user. If the robot finds multiple candidate objects, it first asks the user questions about the object characteristics called attributes that are useful in determining the candidate.Then after getting the information from the user, it refines the candidate set again based on the attribute. Using deep learning, the attributes of each objects is extracted to generate the optimal question to identify which object the user is talking about. We generated a training dataset by using Open Image V5 by Google.

The goal of this research is to develop an efficient object search method by an autonomous mobile robot. There are two main steps in object search. One is the object detection step to detect an object using sensors, and the other is the viewpoint planning step to move to the next viewpoint when the object is not detected. This research focuses on the viewpoint planning step that decides where to move for the next observation. In object search tasks, since searching unknown areas for the target object is preferable, we plan the next viewpoint which maximuize the unknown area to be observed. We apply the viewpoint planning method to the task of searching for a specific object on the desk using a human support robot (HSR) and verify its effectiveness in simulation and real environments.

Recently, the development of housekeeping robots such as robot vacuum cleaners has been actively promoted, and many cheap ones have appeared. But advanced tasks need large-sized products for industrial use are mainstream. In this paper, we develop a system that automatically aligns and arranges shoes by integrating the image recognition of the camera mounted above and the operation by a mobile robot, targeting the arrangement and arrangement of shoes at the entrance. Experimental results show the effectiveness of the method to determine the path for aligning the shoes based on the position and orientation of the shoes estimated by the video recognition of the camera.

Recently, robots that perform housework on behalf of human beings have been developed. However, there are few robots that can do housework related to water. In particular, bathroom cleaning is the place where automation is most desired. In this study, we propose a robot mechanism that cleans a bathtub.

In recent years, life support by robots is expected because aging is advancing. In order to perform object manipulation or action determination, situation recognition function that acquires information from the outside world is firstly required. In particular, when performing situation recognition, visual and auditory senses are very important factors. In this paper, we proposed the audio recognition system focusing on sound lengths and the visual recognition focusing on object states. In addition, we focused on learning the relationship autonomously between what robot is looking and what robot is hearing. We have introduced the system that robots can be looking for the cause when they are hearing an unknown sound, and have realized the life support behavior based on sound.

In this paper, we present a path planning approach for mobile robots using semantic maps. Semantic mapping is qualitative description of the robot’s surroundings, that aims to augment the navigation capabilities and task planning for robots operating in human-robot scenarios. It presents a knowledge layer over an existing map representation such that the knowledge information can be used by robots to learn tasks in a human friendly way. We present a planning scheme using such semantic maps, in order to improve the goal reaching capabilities of the robot in a task-planning scenario. Our method can integrates existing path planners into the topological graph planner that produces faster goal searches.

To resolve the lack of labor caused by the aging population, the social progress of working robots is expected. To realize their social progress, we should analyze and remove the elements of making consumers avoid their social progress. In this paper, we attempt to clarify the differences in consciousness about robots caused by the different backgrounds of home-economics, science, and robotics students, through a lecture and group discussion. From the result, we found that 1) home-economics students tended to have negative emotions such as fear and anxiousness about robots and 2) the lecture was useful for mitigating such negative emotions.

Robots are desired to be a system which not only substitutes the user’s task but also augment the his/her physicality. Human support robots such as a partner robots are needed to have an intuitive interface which does not interfere user’s activity. Therefore, cognitive sharing between users and robots has been a challenging problem for such an interface system. In our research, we develop a cognitive sharing system based on matching of smart glasses and robot vision. In this paper, we realize the navigation system which the robot follows the path the user drawed with hand gesture based on our proposed system.

In order for a home robot to manipulate household objects such as furniture and home appliances, it is necessary to know the internal state of the target object and the interactive relationship with other related objects of its manipulation. We focus on the task of tidying up among multiple house tasks expected to be realized by home robots. In tidying up an office chair, it is necessary to flexible plan a motion generation considering the interactive relationship between office chairs and desks, such as storage areas and combinations. This paper addresses an autonomous motion planning of tidying up a chair considering the positional relationship between a robot, a desk and a chair.

This report explains the development of IoT based standing aid handrail which is the device for human motion sensing. Although it is quite important to maintain healthful daily life of elder people, it is not easy for them to have frequent health examination. This study aims to check their health condition by adding IoT function to facilities used in their daily life. The IoT based standing aid handrail explained in this report has sensors to measure the load given with user’s standing motion and the acquired data is used to estimate muscle weakness of the user. The experiments validate the capability to sense the standing motion. The future expansion of system is also addressed.

Validity of the band-type soft actuator device for feedback of prosthesis hand grasping was verified. This actuator can bend by compressed air and be wrapped around the respective parts of the prosthetic hand user’s body with a force proportional to prosthesis grasping force. Therefore, the actuator bands can provide the prosthesis grasping power information with the feedback. In this study, the soft actuator device was applied and attached to the ankle of examinees, and then recognition rate of grasping power by the actuator was investigated in experiments. In the results, the recognition rate which was high to some extent was confirmed in the tests.

Since 2018, Nagaoka city office has been organizing a senior care innovation hub in order to accelerate IT/IoT/Robot implementation into the care facilities in Nagaoka. In this paper, an overview of the innovation hub is explained, and the activities are examined form the view point of open innovation and local area collaboration.

Falls in the elderly include risk of injuring such as a fracture. Trunk balance exercise is effective to prevent falls in them, virtual reality (VR) intervention for exercise increases the motivation of the participant. This study aimed to develop the VR trunk balance exercise device in sitting. The device was produces that it can improve the balance ability of participant by game-based trunk balance exercise in a virtual environment. It also feedbacked head motion of them by the user interface and showed a 360-degree video. In the experiment for healthy adults, the amount of head motion in the lateral direction while exercise is decreased significantly by function feedbacking head motion, the 360-degree video was increased their satisfaction without the effect to motion sickness. VR trunk balance exercise device is safe and effective.

In this study, an electric assist suit which can be used in both walking assist for elderly people and lifting motion in transfer work for caregiver is discussed. First, the prototype of electric assist suit which was developed in our previous study was improved so as to reduce the weight. As a result, the assist suit became 4kg lighter compared with before, so that various people can wear the improved assist suit. Then, a new approach to control walking assist was proposed, in which characteristics of individual walk were taken into consideration. To demonstrate the validity of the proposed control method for walking assist, the evaluation experiment was executed, and validity of the proposed method was verified through experimental result. Finally, usefulness of the improved assist suit was shown through the verification experiment of the assist suit.

In this study, prototype of an upper arm assist suit for upward work such as harvest of fruit and ceiling covering at the construction site was developed. In order to evaluate the developed upper arm assist suit for upward work, three comparative experiments were executed for four subjects under the conditions of wearing and unwearing the assist suit. In those experiments, measurement of electromyogram, measurement of muscle hardness, and VAS evaluation were conducted. From the experimental results, reduction of the burden of shoulder during upward work was observed. Thus, the effectiveness of the developed upper arm assist suit for the upward work was verified.

In our article, we propose a Body Weight Support Walker called FLORA TENDER. The user wears a harness which is connected to the walker by two wires and the walker pulls up the patient. However, when doing manual work while moving with it, the user might not be able to go toward the intended direction. We propose a method for controlling the orientation of it’s caster wheels, so that it easily moves toward the intended direction by the user. In this research, a user state estimation method based on the angle and the length of the wires connecting the user and the walker is proposed. In addition, by steering the caster wheels towards the appropriate direction based on the estimated state, we enable the FLORA TENDER to move easily and consistently along with the user’s motion.

Rehabilitation and training must be performed in an appropriate posture, and if performed in an inappropriate posture, the target muscle may not be effectively stimulated. Furthermore, depending on the posture, there is a risk of applying unnecessary load to specific muscles and tendons. Thus, it is known that effective stimulation of the target muscle depends not only on the control of the training support system but also on the posture during training. The purpose of this study is to measure trainees and evaluate their posture so that they can perform effective training with proper posture during lower limb muscle strength training. In addition, it is necessary to evaluate the posture automatically because a skilled person does not always accompany the training.

With an increase in lifestyle-related diseases, the number of patients with deformed feet due to vasogenic lesions such as diabetes has increased. In particular, prostheses are required to adapted to their own body shapes due to foot loss. However, there is a problem that the material used for the production of the current prostheses requires a certain period of time to cure. The purpose of this study is to apply the jamming transition phenomena and to develop an element with stiffness change characteristics at any time. In this study, a prototype of a jamming element is manufactured, and the shape reproducibility is examined to clarify the sampling characteristics.

When wheelchair users board a train, they require a station staff’s assistance because there are step-difference and gap between station platform and train. Therefore, they feel stress for the movement with the train. On the other hand, the labor burdens on station employee increase with increase of the use of station by the wheelchair user. In this study, an assistance device is proposed to solve such problems. This device is available for an existing wheelchair and a platform and a train and does not need power such as the electricity. The wheelchair user can board a train by oneself without needing assistance by using this device.

We have been studying a support robot teaching method to grasp a various shape object in our living space by clicking with a TOF laser pointer. But because the selection of gripping object used for the experiment was ad hoc, the criterion of the grasp-ability for any shape object was unclear. As a result, types of grippable objects had significant limitations. In this paper, as a first step to clarify the criteria, we focused on cuboid as gripping shape, performed comprehensive grouping with respect to three side length ratio, proposed grasp teaching method applicable to each group, and demonstrated its effectiveness.

In this paper we propose a steering control method for assisting in using the cycling wheelchair while focusing on maintaining the user’s sense of agency. The purpose of this method is to support users to move by themselves when they are shopping in a wheelchair, which is one of common scenes when going outdoor. Our method is to determine the assist ratio based on environmental information and the force applied by users. When the assist ratio is settled within the range we defined, we support steering based on environmental information. On the other hand, when it is out of range, we limit the output to prevent excessive assistances. We verified the effectiveness of the proposed method focusing on user’s sense of agency through experiments The result shows that considering environmental information and the force of user’s operation might be related with better traveling performance while maintaining user’s sense of agency.

In this research we propose a novel mechatronic Body Weight Support (BWS) system for locomotor rehabilitation training in patients with stroke or spinal cord injury (SCI). Body Weight Support Training (BWST), which has been proved to be a promising therapeutic tool to relive patient’s body weight and therefore facilitate gait pattern. We utilize Variable Stiffness Mechanism (VSM) for the system mechanical design to control the virtual stiffness of BWS system and thereby support human body weight. Experimental result shows the effectiveness of both the hardware design and control method to alleviate the body weight during walking motion.

The purpose of this research is to automatically diagnose the risk of frailty of elderly people by using wearable gyro and acceleration sensors in daily life. We focus on the difference of sit-to-stand motion strategies. In the proposed method, sensors are attached to the chest and thighs of the person to be diagnosed. They are asked to perform two types of standing twice. The anteversion time, standing time, trunk anteversion angle, average angular velocity, and instantaneous angular velocity are picked up as features. Influence of each feature is viewed using logistic regression models. Experiments were conducted with 53 subjects. As a result, there was no significant difference between healthy and frailty groups. The reason for incorrect diagnosis would be because the frailty subjects were the persons who tend to have social frailty rather than physical frailty. The problem of more appropriate experimentation design is left for further research.

We propose a new a lightweight and compact single-wheeled pendulum-type robot to assist rehabilitation using light-touch effect. By applying the light touch effect, walking is stabilized by reduction of swing of center of the gravity. In addition, we examined the use of multiple sensors to determine external situations, estimate the user's intention, and make operational announcements. Then, it was verified walking stabilization by two cane robots. As a result of the experiment, it was confirmed that the center of gravity swing is more stable than when walking with two cane robots using one cane robot.

In this research, we construct a lightweight and highly reliable flight reservation system of autonomous distributed unmanned aerial vehicle traffic management system (UTM) using Blockchain technology on a small PC such as Raspberry Pi. The current UTM system is almost centralized, but it needs so much data size should the system control which is calculated by NEDO DRESS PROJECT, and it is quite difficult to implement in real. Our flight reservation system will be autonomously decentralized using Blockchain technology. We use a computer language called Rust, manage and index the airspace using OcTree on P2P, and locally manage "purchase airspace" with Blockchain technology as a ledger. By making consensus building between the nodes that do so, reservations can be made.

In this paper, the system configuration, data transmission contents, and received result contents of robot networks are shown for the experiments using general-purpose communication units at iREX 2019. By using the RSNP unit, a wide variety of robots, regardless of OS and middleware, could be easily networked and their data were collected. Here, we explain the proposed system, connection method, data information and experimental results. Next, in addition to collecting data, we will plan to use the data to provide feedback to other robots and users.

Recently, robots get more DOF (Degree of Freedom) s and there are variety of motions or control approaches for them. Thus, there may remain motions for that robots beyond the scope of developer ’s assumption. In this research, a public interface in network for collecting robot operation information is proposed. This system is expected to collect and analyze user’s operation data to generate new motions or improve conventional motions. This idea is applied for a simulator for walking motion of quadruped robot considering stability. An interface which provides graphic display, operation interface and stability calculation is developed on web browser using HTML and JavaScript and its function is verified.

We are developing a ROS-based data-driven autonomous mobile robot for research on robot systems that involve both autonomous movement and teleoperation by the user. We conducted verification experiments from data collection by teleoperation by the user to autonomous movement.

We have been discussing a disaster information gathering system using mobile robot. There is a problem that an accident occurs on the robot in case of disconnecting the wireless LAN between the operator and the robot. Then, we proposed wireless mobile robot tele-operation system using LoRaWAN as sub-communicator to solve the accident. In a wireless LAN tele-operation, the operator operates the mobile robot by using picture or movie information from the robot. However, in a tele-operation using LoRaWAN, operator can’t get that information because of the low transmission rate. Then, we proposed an image transmission method using spread spectrum of LoRaWAN communication. In experimental results, we confirmed that the proposed method is effective for the tele-operation of mobile robot.

In this study, we have discussed communication quality measurement method between an operator and the mobile robot for tele-operation of mobile robot in the information gathering activity in the closed space. In tele-operation of mobile robot, operator sends operation command packets at regular intervals. If packet arrival interval is delayed, it affects tele-operation and video streaming. Therefore, when tele-operation of mobile robot, it is necessary to measure jitter and grasp the delay of tele-operation. Hence, we proposed jitter monitoring method based on packet arrival interval in low load throughput measurement. In this paper, we evaluated the proposed method, then it is confirmed that validity and effectiveness of measurement results.

We have been discussing a disaster information gathering system using mobile robot. Leaky CoaXial cable radiates radio waves inside the cable to the outside, therefore it is possible to operate mobile robot along the cable. There is a problem that Leaky CoaXial cable is disconnected in disaster area. In this paper, to settle issue with accident of mobile robot due to disconnection of Leaky CoaXial cable, constructing reserve communication infrastructure is proposed. The proposed method is a method to return the mobile robot to the operator's place via reserve communication path that relayed the calculator. The validity of the proposal method was inspected by doing an evaluation experiment of mobile robot tele-operation.

This research aims to generate a feature-based topological map that extracts and stores features of a place using an omnidirectional camera as a means of navigation. A specific place is recognized by using YOLO object detection and Optical Flow results. The closed-loop of the map is detected by a PCA-sift image feature extraction algorithm and the feature comparison using a fast k-means clustering algorithm (AFK-mc2). Furthermore, a topological map connecting each node is created. The proposed method is expected for navigation under limited memory space.

For robot applications, its task performance and operation guarantee are important. Usually, there is a trade-off between generalization ability and stability for a target task. Deep learning is a promising approach to complex tasks, however, it is difficult to deal with motions that do not include in the training dataset. In this study, we take the approach of guaranteeing that undefined motion by model-based method. Our method switches between model-based controller and a learning-based controller from the internal representation of Recurrent Neural Network. By using our method, the error recovery of the robot task can be realized. We evaluated our method by using a real robot, Torobo ARM.

We propose a deep learning model to select and handle appropriate tools in cooking scene. The tool to be selected depends on both extrinsic characteristics such as size and shape, and intrinsic characteristics such as hardness and viscosity of the ingredients. However, in previous research, they were difficult to use tools in consideration of the characteristics. We allow a robot to stir the ingredients and conduct “active perception” to recognize the characteristics by itself. Then the robot generates motions to move the ingredients from a pot to a bowl with a spatula or a ladle using the deep learning model. In evaluation experiment, we allow the robot to operate untrained ingredients, and confirmed the generalization ability of the deep learning model.

No matter their environment, it is important for robots to ensure safety and prevent malfunctions. An important aspect to robotics is therefore collision-free path planning. Conventional methods such as RRT and potential field have difficulty adapting to dynamically changing environments, and have high calculation cost for iterating or searching. In this research, we propose a new path planning method in which the latent space of Conditional Generative Adversarial Networks (cGANs) represents the workspace where the robot avoids obstacles, and a path is planned in this latent space. The selected path is then mapped to a path in joint space. Since the latent space is collision-free, path planning can be performed by connecting the start and the goal robot posture without iteration or searching processes. We confirmed that our single trained model can handle various types of multiple untrained obstacle conditions.

The motor directional tuning phenomenon that can be seen in the brain of primates suggests that the arm’s movement direction is one of the most important parameters in reaching movements. In this research, we have created a controller based on this phenomenon, and applied it to a musculoskeletal robot arm. The controller does not have an internal model, and provides a mapping from movement direction to muscle commands. Despite a relatively simple control scheme, the controller has achieved performance comparable to a model-based controller, and achieved efficient usage of the complex musculoskeletal robot body.

Though there are a lot of researches about Physical Human-robot Interaction (pHRI) using prediction, few researches work on inducing the opponent’s action or outwitting the opponent. We made the push-hand game environment in order to focus on generating strategic actions and tried to make reinforcement learning agents to learn these actions by adding rewards which are directly proportional to the degree of inducement (induction reward) or the degree of outwitting (outwitting reward), defined in this research. As a result, we demonstrated that the induction reward decreases the agent’s predictive error and the outwitting reward increases the opponent’s predictive error, and both of them didn’t contribute to the winning percentage.

This paper proposes a novel variational autoencoder derived from Tsallis statistics, named q-VAE. Starting from the viewpoint of Tsallis statistics, a new lower bound of the q-VAE is derived to maximize likelihood of the data sampled, which has a potential for disentangled representation learning. As another advantage of the q-VAE, it does not require independency between the data. The q-VAE is demonstrated in learning latent dynamics of a nonlinear dynamical simulation. As a result, the q-VAE achieves stable and accurate long-term state prediction from the initial state and the actions at respective times.

The purpose of this research is to design a throwing motion of 3-DOF manipulator so that the thrown object lands on the target position. For this purpose, (1) a physical model have to be identified to calculate an appropriate feedforward torque, and (2) the motion which minimizes sensitivity with respect to parameter error is designed. In (1), identification is conducted to identify minimum dynamic parameters of the manipulator. In (2), because the identified parameters have uncertainty due to a noise etc., the error covariance of parameters will be identified, and a throwing motion is designed to minimize sensitivity of landing position error considering the obtained error covariance. Finally, the effectiveness of the proposed method is verified by experiments.

Actuators that have characteristics such as light weight, high torque to weight ratio, and torquecontrolled performance are desired for finger robots or power assist suits. Although electromagnetic motors excel in torque-controlled performance, they are low torque to weight ratio. On the one hand, although electromagnetic motors with reduction gear are high torque to weight ratio, they are regarded as low torque-controlled performance by backlash and friction inherent in the actuators. This research therefore design load torque controller using motor/load-side encoders and load torque estimating observer using the backlash model. The validity of the proposed controller and observer are verified by simulation and experiment.

When the operator moves the robot arm (master), a propeller driven system (slave) such as a multi-rotor helicopter tracks the position, and when the slave comes into contact with an object, the contact force is feed back to the operator. (Below, the bilateral control systems) are attracting attention. Switching controllers that switch between controllers when contacting and non-contacting with an object have been proposed, but are susceptible to modeling errors in the contact object’s stiffness. In this research, we propose an adaptive controller that estimates the contact object’s stiffness in real time and uses it for control gain. The validity of the proposed adaptive controller is shown by experiments.

In recent years, dynamical systems handled by machine learning become remarkably larger and more complex, but they require intractable costs due to high-dimensional raw observation space. Variational autoencoder (VAE) is one of the promising approaches to the problems since it has a capability to learn dynamical systems in latent state space hidden in data, which can reconstruct system characteristics. Until now, however, VAE has been applied only to prediction and/or anomaly detection. This study, therefore, applies VAE to control application task in the latent space. Specifically, iterative Linear Quadratic Regulator (iLQR), which is one of the model predictive control methods, is integrated to VAE. In a benchmark problem, our method achieved the desired motions even using the latent space dynamics.

This study proposed the method for suppressing residual vibration occurred in a grasping object by optimizing the trajectory which determines a change of the displacement and posture of a robot with time. In the proposed method, the residual vibration measured by actually operating the robot is evaluated, and the trajectory is optimized through the heuristic algorithm. Therefore, a robotic dynamic model that is generally difficult to construct is not required. As a result of applying the proposed method to a commercially industrial robot arm operating along a linear locus, the optimal trajectory generated by the heuristic algorithm succeeded in reducing the settling time of the residual vibration by 98 % compared to the manufacturer’s standard trajectory.

In this paper, we propose a novel conveyor system that achieves transportation on low friction surface by utilizing our crawling-like locomotion robot which owes the capability to slide on slippery ground．First, we derive the equations of system dynamics and control．Second, we analyze the effect of system parameters on the velocity of transported object using this method via numerical simulation. Finally，we design an experimental machine for further verification of the numerical results．