The purpose of this study is to develop a four-wheeled autonomous mobile robot for transporting heavy loads on a narrow alley of 1380mm width with a steep 20 ° slope and right-angle curves. The robot is driven by four wheels with hub motors in order to climb a steep slope with maximum load. In addition, for high-precision autonomous driving, one LiDAR is equipped on the front and rear of the vehicle respectively to acquire environmental information, and a self-positioning algorithm is developed. Further, verification experiments are conducted on two courses that simulate real conditions. As a result, we confirmed that the vehicle was able to climb steep slopes and drive autonomously with high accuracy on narrow alleys.

In order to enable adaptive mobility in debris-strewn environments such as disaster sites, swarm mobile robots with the ability to connect and separate multiple units. In this study, a unit with spherical outer shells is proposed. To reduce the size and lower the position of COG of the vehicle, actuators, a small coupler, a small microcomputer, and an auxiliary wheel are placed using internal gears inside the spherical outer shell. To shorten the fabrication time and facilitate mass production of the vehicle, the mechanical design and manufacturing presupposes the use of 3D printers. In addition, wireless control system for multiple units are developed. The performance of the developed vehicle is experimented by connecting/separating, running in a ditch, running over steps, and running in a narrow space.

This article proposes a new positioning mechanism for a mobile robot, Mechanical Positioned Guided Vehicle “MPGV”. The MPGV uses a clutch and electric stoppers to mechanically interrupt driving torque of its motor and stop. Compared to general AGVs, the mechanical method allows mobile robots to be control-less and sensor-less. The experimental results indicated that the MPGV has a positioning accuracy comparable to that of general AGVs.

This study introduces an action model for mobile robots that follows the specific walking human ahead of the robot and reaches destinations with the target human. The action model is trained by deep reinforcement learning in simulation environments including the target human. A conventional study of the authors shows that the training using 2D-LiDAR scan can provide action models for navigation to the destinations and following human flows. In this study, the robot distinguishes humans and the other static obstacles and distances to the human observed by the robot in addition to scan data are adopted as input state of the training. As the results, following of the specific human was achieved in the simple simulation environments.

The purpose of this study is to improve the traversability of the combined articulated crawler robot. We propose a new mechanism that enables the sub-arm to move in a position/force-tracking using a planetary gear mechanism. The output shaft torque enables Reduction of installed motors, Position/Force-tracking of a Sub-Arm, and Active movement of the Sub-Arm depending on the output shaft torque relationship. The purpose of this study is to improve the traversability of the combined articulated crawler robot. We propose a new mechanism that enables the sub-arm to move in a position/force-tracking using a planetary gear mechanism. The output shaft torque enables Reduction of installed motors, Position/Force-tracking of a Sub-Arm, and Active movement of the Sub-Arm depending on the output shaft torque relationship.

In recent years, with progress of automation in logistic center, AMRs(Autonomous Mobile Robots) are introduced to solve labor shortage and improve work efficiency. We are developing AMR which can be applied for roll box pallet transportation and alignment in logistic center. The utilization of mecanum wheels makes AMR omnidirectional and AMR is able to move freely in narrow space. However, when AMR transports roll box pallet, for safety, AMR keeps casters of roll box pallet on the ground. Which makes AMR has to follow two differential wheel movement of carried roll box pallet. In this paper, we propose a method of planning trajectory for AMR under the constraint of two differential wheel movement while aligning roll box pallet.

This paper introduces deep reinforcement learning to obtain action models for autonomous mobile robots in the simulation environment. The proposed robot system uses monocular camera images mounted on the mobile robot instead of 2D-LiDAR. Unity to create simulation environments and agents was adopted in this study. The training in Unity provides action models that the agents can reach the destinations in the environments using monocular camera images input. In addition, the trained models acquired in the simulation environment were adopted to a ROS-based navigation system in Unity.

Currently, there is a wide range of developments in the area of robot mobility. In the disaster areas, where it is difficult for humans to walk, and especially in not flat place, various methods and experiments have been studied from various viewpoints. On the other hand, there are relatively few studies on the snow environment, where mobile mechanisms may easily get stuck by snow damage. The purpose of this research is to develop mechanisms for a snow mobile robot that can move stably for investigation and carrying of emergency supply at the place suffering for snow damage. This study focuses on the development of artificial wheels which give both buoyancy and propulsion force against snow. Also, this study analyzes physical characteristics required for experiments in snow environments and proposes artificial pseudo-snow field for in-door experiments.

For wheeled mobile robots with non-holonomic constraints, it is difficult to derive a control input that the robots are convergent to the target state. In this paper, we consider a regulator problem of controlling a two-wheeled mobile robot with non-holonomic constraints from an arbitrary initial state to a target state, and propose a method to generate a path and feed forward control inputs to converge the robot to the origin using time-polynomial. The effectiveness of the searched control input is confirmed by numerical simulations.

In this paper, in an inverted pendulum type stair climbing robot, which has a mechanism that rotates two-wheeled rotating links and moves the center of gravity with a slider at the top, we applied a rack and pinion mechanism to the rotating links and proposed a mechanism that moves the rotating links in parallel to move the center of gravity. As a result of the control experiment of the experimental system, it was found that it was possible to control the rotational motion and parallel movement of the rotating link. The load torque applied to the gears for rotation and translation was measured in order to select a motor to control the links.

The Mars Exploration Rover will travel over the surface of Mars to acquire detailed topographic data and collect samples. However, the Martian surface is covered with fine sand called regolith, which prevents the rover from following the target path accurately and causes it to get stuck. Therefore, the rover must have high performance on rough terrain. In addition, the rover must be lightweight because it is transported by a rocket, and the mechanism must be simple to prevent failure due to regolith. Therefore, a passive steering mechanism using springs is proposed, and the effectiveness of the proposed method is confirmed by driving experiments.

We are developing a towed underwater robot and conducting a demonstration experiment to create a 3D map of the seafloor. From the operational issues, we will try to reduce the size and weight of the towed underwater robot. In this presentation, we describe the design of a towed underwater robot that is smaller and lighter.

Low-cost AUVs (Autonomous Underwater Vehicles) are effective in improving the efficiency of ocean exploration, but their poor sensor performance makes positioning a challenge. In order to make positioning more accurate, we developed a method to estimate AUV heading error and ocean current. In this article, we report the results estimated by post-processing using the experimental data.

AUVs (Autonomous Underwater Vehicles) for underwater research are attracting attention for their practical use in the fields of research and ocean development. We found some problems with the AUV created in 2015, so we improved it. One of the problems was the reliability of the circuitry, which had been created by us using NC machining, so we redesigned the circuitry and outsourced the circuit board creation to ensure reliability. The circuits we created were a motor circuit and an encoder circuit, and we checked the operation of these circuits.

The purpose of this research is to develop a small haptic control device for operation support of underwater robots. We used a flight joystick for haptic shared control of underwater robots in our previous work. However, the flight controller is not suitable when we use it on small ships that are unstable. To solve this problem, we designed and developed a small haptic control device. The developed controller can be operated by both thumbs to provide input commands for three-dimensional robot motion. This controller can also display force to a robot operator. We conducted experiments to verify that the developed controller can be used for haptic shared control of an underwater robot.

AUVs (autonomous underwater vehicles) for underwater research are attracting attention for practical use in the fields of research and marine development. We found a problem with the AUV created in 2015 and made improvements. The problem was that water intruded into the PC part inside the hull and the friction loss of the magnet coupling of the thruster was large, so we changed the hull to two by external design, and made the thruster lid by ourselves and made it waterproof. did.

In this paper, we consider the application of haptic shard control to an floating underwater robot for surface sediment removal. First, we experimentally measured water flow velocities generated by marine thrusters to determine the thruster arrangement for surface sediment removal. Based on the thruster arrangement, we discuss a method to compensate for the reaction force acting on the underwater robot during sediment removal.

In order to prevent the tether cable from adversely affecting the ROV’s motion performance and cable entanglement during ROV operation, the cable must be properly routed according to the direction and depth of the ROV in the water. In this study, we propose a motion plan for the position of the ASV and the cable length with the motion of the ROV in the ASV-ROV cooperative system.

The prototype AUV(Autonomous Underwater Vehicle) has a propulsion mechanism that rotates the sealed outer shell, but there was a problem that the internal drive and control unit also rotated due to the viscous resistance of the water and the reaction torque. Therefore, the purpose of this paper is to propose and verify a method to suppress the rotation of the drive control unit due to the reaction torque. The proposed method is based on the assumption that the counter-torque of the drive control unit is offset by the rotational torque of the flywheel. In order to prove this assumption, we conducted a control experiment in which the flywheel motor rotates at the same angular velocity as the drive control unit. As a result of the experiment, it was confirmed that the flywheel motor can follow the rotating drive control unit.

Sperm whales are known to prey on mesopelagic cephalopods such as giant squid. However, their predatory behavior has not yet been recorded as photographs or videos. If such images can be captured, it will be a valuable clue to clarify the predatory behavior. Therefore, we are developing the “whale rover” that is attached to the back of the whale and moves to the vicinity of the mouth. In this paper, the improvement of the whale rover and the results of field experiments are described.

The ocean is one of extreme environments, and underwater robots are expected as the tools for observation of the ocean. Underwater robots require high autonomy and are self-contained to accomplish various deep-ocean observation missions. The next generation AUV is expected to have sampling function of marine benthos with end-effector to catch and bring back marine creatures. However, sampling is not easy task for AUV as the shape, size, motion of targets differs. So that the end-effectors is required to have enough accuracy in control manipulation. One of the key issues is the target recognition from underwater images. In this research, floating objects cancellation method from underwater images using optical flow is discussed.

Currently, underwater infrastructure facilities in Japan are rapidly aging, but most of them are not inspected by divers because the current inspection by divers is inefficient and potentially dangerous. To realize a system that can inspect underwater infrastructures safely and efficiently, this study develops a motion control system for an underwater robot that is part of the underwater infrastructure inspection system.

Optimizing fish feeding is needed for a sustainable marine aquaculture production. We approach this problem by estimating the behavior of fishes during feeding by measuring currents they generate around the cage. In this paper, we present our development of an underwater sensor network consisting of current sensors, inertial measurement units, and cameras. This system was deployed in a fish cage and recorded measurements in three fish cages during feeding activities. Results showed that currents generated by fishes were largest at the surface during feeding and decline over time until the feeding ended. Further data analysis is needed for optimizing the farmer’s feeding decisions.

Proper communication between a patient and their caregiver is paramount to insuring proper medical treatment. Speech disabilities and certain medical operations may hinder this communication and can cause issues for all parties involved. We have developed a wearable device, named the alternative communication glove (ACG), that allows a user to type any Japanese character using only two independent finger movements. We propose this unique methodology to compensate for a niche group of users that may be overlooked by current products and research. The main benefit of the ACG is that it can allow the user to express themselves while in any body position and without the need for eyesight.

For the safety of elderly people, it is necessary to assess their balance ability. To assess the balance ability, the stabilometry with a force plate, which measures the trajectory of Center of Pressure (CoP), has been used in clinical fields. Although the force plate can precisely measure the trajectory of CoP, it is expensive and bulky. To address this problem, the authors propose a method to estimate the trajectory of CoP using an internal measurement unit. In this study, the trajectory of CoP during human standing was estimated using the proposed method and the estimated trajectory of CoP was compared with the CoP measured by force plate. By comparing estimate the trajectory of CoP to measured CoP by a force plate, it was found that the trajectory of CoP estimated by the proposed method represents the features of the trajectories of CoP during human standings.

Office work involves prolonged sitting posture, which causes back pain. Various methods have been proposed for the back pain. However, most of these methods have not been evaluated quantitatively for their effectiveness for the back pain. Therefore, this study develops an active chair and evaluates its effecttiveness quantitatively on the view point of activity of trunk muscle which decrease the back pain problem. It is shown that the chair promotes the activity of trunk muscle and the use of the chair does not affect the efficiency of office works.

In this paper, we examined the comfort of common spaces in wards and rooms and living rooms of welfare facilities for the purpose of inpatient treatment and inpatient care in medical and welfare facilities. In order to build an environment where daily life in a common space such as a ward, sickroom, or living room can be spent more comfortably and with peace of mind, (1) when medical staff use the daily living environment in this space, (2) patients and use When people use it mainly, (3) It is assumed that visitors such as visitors come and go. By making the surrounding environment of the user in each situation more convenient and easy to use, comfort is inevitably smoothed in that environment. In this way, we propose the idea of a facility user surrounding environment smoothing system that can provide a living space that satisfies not only facility users but also workers and visitors in total.

Wearable gait assist robots are expected to improve the quality of life of the elderly, and some robots have been developed. However, most robots are set to walk straight and it is difficult to perform curving gait. The purpose of this study is to develop the joint assist pattern suitable for curving corner with wearable assist robots. It is considered effective to adjust the length of time and torque of the robot’s actuator. Experiments were conducted with robots turn the corner to verify the changes in gait parameters. The results confirmed that there was a significant change in walking speed during curving.

In this research, we propose a novel autonomous mobile robot for transporting passive welfare objects in medical facilities. To actuate the robot, together with a welfare object, a differential wheeled robot is developed. Additionally, to combine the robot with various types of welfare objects safely, the robot can adjust its geometrical shape in the vertical direction by using a mechanism called zip chain actuator. When considering the task of transporting a welfare object, the robot must localize itself properly and detect the environmental information continuously. Furthermore, the robot must be able to handle the payload of such combined objects. To achieve this, we implement a Move Base Interface, where various kinds of sensory information including the data of IMU, wheel encoder and laser scanners are fused by using a localization algorithm called Adaptive Monte Carlo Localization (AMCL). In addition, the direction towards the robot moves is reversed during the transportation task, so that it is possible to pull the object and maintain localization simultaneously. Through experimental validation, we confirmed that the proposed mobile robot can follow the local planner trajectory ideally and reached the destination regardless the partial occlusion and unknow dynamics of passive objects.

The complexity of Multi-Agent Reinforcement Learning (MARL) problems increases exponentially with the number of agents. Poor scalability to the number results in limited applications of MARL to large-scale multi-agent systems.

In this paper, we present a novel MARL algorithm leveraging a self-policy network to estimate the intentions of other agents.The intention of other agents is backpropagated from a self-policy network with the observed action of others. Estimated intentions are then used as input to the self-policy network. As long as the agents are cooperative, our method does not require any additional model to learn others’ intentions. We also introduce a simple curriculum learning, which gradually increases the number of agents. Simulation results indicated that the proposed method improves the performance of learned policy even if the number of agents increases.

A joystick control algorithm for a cooperative object positioning system with three manipulators has been developed. The object is suspended by six wires each two of which are fixed to three manipulator tips respectively and is controlled by the three manipulator motions. Since the system has twelve kinematic redundancy, a way to use them to keep the minimum wire tensions and also to avoid the concentration of the load on a single wire is proposed, and the effectiveness of the system has been verified by some fundamental experiments..

In this paper, we propose a cooperative localization method for multi － robots based on the confidence evaluation of positioning. In conventional methods, robots are often separated into two groups: parent robots and child robots. Since child robots depend on the information of parent robots, the error of parent robots will spread to the whole robot system. Therefore, we propose a method to estimate the position by dynamically switching the reference robots without specifying a specific parent robot. Results show that our method can maintain the accuracy of position estimation for the entire robot system better than when parent robots are fixed.

The human brain has surprising plasticity of recovery from stroke. Therefore, neurorehabilitation using apparatus that applies robotic and virtual reality technology is effective for stroke patients. In this report, I introduce rehabilitation systems“New PLEMO” series for upper limbs using virtual reality technology. Rehabilitation systems using brakes are essentially safe unlike rehabilitation systems using actuators.

The purpose of this research is to develop and control an electric wheelchair with high adaptation to driving environment. For this purpose, we propose the concept of the wheelchair to be developed. We also design a control system of the wheelchair. Its power train consists of three wheels on one-side in trident configuration and planetary gear mechanism. An experimental verification of the power train has been performed.

When a passenger surrenders himself/herself to a wheelchair that is hand-pushed by other’s hands or automatically operated, he/she may feel uneasy, for example, near obstacles or steps. The purpose of this study is to clarify the relationship between the passengers feel discomfort due to external factors, the speed of the wheelchair, and the distance to obstacles. For this purpose, an experimental robot based on an electric wheelchair with an sensor function to mesure the environment. The speed and distance from the wall were varied by the device, and We quantitatively measured the discomfort level of the subjects in each situation. This describes the experimental method, and the results of the analysis.

This is study on control method of power assist suit that supports lifting motion and reduces the load on the waist of the wearer. The purpose of this study is to determine and control the required assist force of power assist suit from the parameters obtained by the sensor. To control magnitude of assist power according to the posture of wearer, we made dynamics model of lifting body and calculate load on the body. We used linear support vector machines to classify the amount of weight of a luggage.

In this research, we propose a novel lightweight and compact walking-assist robot to prevent users from falling accidents. We propose the concept for a new walking assist robot. A two-legged wheeled balancing robot is developed to follow the user’s walking motion based on the concept. Each leg consists of a two-link structure, including one driving wheel and two revolute joints. A base frame was connected by the base of both links. Three basic controllers were developed to keep the base frame at a specific angle, change the robot’s overall height and keep the robot’s balance for moving. We utilize a motion capture system called VIVE to keep the specific distance between humans and robots. The validation experiment shows that the proposed robot can follow user’s walking while keeping balance simultaneously.

In order to clean a floor efficiently using large and small robots simultaneously, it is useful to operate each robot in a planned manner. In this study, we compared the efficiency of cleaning using a path finding method and a action decision method according to the state, and presented that there is a suitable method depending on the environment. In addition, we established an efficient cleaning method under realistic conditions by using the action decision method according to the state, and verified the efficient cleaning method by experiments.

In recent years, there are a lot of development and research on automatic conveyance due to concerns about the decrease in the working population. However, conventional research has problems such as route selection is limited due to the large turning radius and the inability to pass through narrow aisles. To solve these problems, we propose coordinated conveyance system using multiple two-wheeled robots. This system can move a transport cart in all directions by controlling multiple two-wheeled robots with an active caster control model so the issues can be solved. In this study, we considered and discussed algorithms for forward kinematics calculation by modeling using virtual link models, detection of uncoordinated states using the change in virtual link length as a flag and updating kinematic models when the number of robot connections changes, which are necessary for controlling this system. We also verified these algorithms using a simulator developed in C++.

This research proposed to develop a rover system that follows a human and runs in a platoon for the disaster response. In the case of this system, we need non-collision route planning avoidance actions to prevent collisions and know the route accurately. In this paper, we developed a system that uses a flocking algorithm to prevent collisions and ROS Topic from sending the front rover or human’s position, velocity, and angle information. At the same time, we search the appropriate frequency for following and platooning in the simulation test. As a result, we realize a rover-following and running in a platoon with four or more rovers and automatically return to the starting point with three or more rovers in the simulation test. In an actual machine test, we realize a human following and running in a platoon with two types of rovers and return to the starting point automatically with a rover. In addition, we find the appropriate frequency. As a result, the frequency is about 5Hz.

In gait analysis, ground reaction forces (GRFs) provide important information about gait function assessment. However, due to the financial and time-consuming costs, the use of force plates for gait assessment in the rehabilitation field is limited. To solve these problems, previous studies have conducted to estimate the GRFs without the force plate. The purpose of this study is to estimate the 3-axis GRFs during walking from kinematic data of markerless motion capture system. For this purpose, we proposed a GRF estimation model using LSTM. The accuracy of the proposed model was verified by comparing with the estimation model using kinematic data obtained from marker-based motion capture system.The proposed model was found to have an estimation error of 20% compared to the measurements obtained from the force plates.

Myoelectric moments are calculated by the positions of electrodes weighted by myoelectric signals. The quantities are helpful in discriminating motions. We have already investigated the myoelectric signals from the hand shapes and the wrist motion. However, the discrimination of finger motions is more challenging because the finger motion generates smaller myoelectric signals than wrist and hand motions. In this paper, we preliminary investigate the myoelectric moments for various finger bending.

One of the reason why elderly people should be nursing is falling. It is important to prevent falls in the elderly. It is known that improving balance is a good way to decrease the risk of falling. Wearable Balance Training Device was developed and validated for it can improve balance ability. In this research, training target is reactive postural control known as balance function related to falls. Subjects were asked to conduct reactive balance test. The result of test was compared before and after training that device cause perturbation in the left and right directions. Reactive balance test showed an improvement in balance in the left and right directions.

In aged society, it has been a serious problem that demand of the assistive technology has been complicated. Although the type and strategy of the assistive technology for gait motion has been expanded, the assistive technology which suits for hemiplegia patients in their daily life has some troubles. As one of the solutions for this problem, we have proposed a walking assist system by using the transmission of the motion energy from unaffected hip joint of the user without external power sources. The key component of the proposed mechanism is energy charge from the unaffected side and use it as the assist torque for the affected side at the suitable timing. In this paper, the design of the proposed mechanism which consists of the energy charge and assist system has been reported. In addition, wearability of the proposed mechanism for subject people have also considered through the questionnaire results collected in our previous experiments.

Training time is shorter because the number of physiotherapists (PT) is not enough compared to the number of patients. This will reduce the effectiveness of walk-training. Patients do walk-training at home rather than on a walker. In this study, we propose a walking motion measurement method using an RGB-D camera when a human walk with a walker, which is for the purpose of enhancing the rehabilitation effect. An RGB-D camera is attached to a care-walker. The RGBD camera measures the position of the feet to estimate walking movement. In this article the proposed measurement method was improved in order to solve the problem of not being able to see the foot when walking and rotating. In addition, we perform a TUG gait test with a pseudo-patient and measure and analyze gait information.

One of the main treatments for gait disorders after stroke is an ankle joint orthosis. Although the spring stiffness of an ankle orthosis affects the movement of the ankle and knee joints, a simple system for selecting the optimal spring stiffness to improve the gait of individual patients has not yet been proposed. Therefore, in this study, a system that can select the patient's optimal spring stiffness based on walking without orthotics was proposed. Fifteen stroke patients wore the brace and walked in front of the camera. As a result, we were able to estimate the optimal spring stiffness with a high accuracy of 90%.

With the aging of Japan’s population, there is a growing need for daily life support for the elderly. In addition, from the perspective of reducing the burden on medical care workers, automation of household chores by life support robots has been expected. The authors have focused on garbage cleaning, which is one of the most burdensome household chores. In this paper, the authors propose a method to identify garbage by user’s finger pointing because it is too difficult to identify what is trash with current machine learning technology. The proposed method calibrates the finger pointing direction using point cloud processing and image processing with a skeletal detection library. Using the proposed method, a life support robot can identify an object on the tabletop even if the user’s finger isn’t completely straight Elderly people may be difficult to do so. As a result of pilot experiments, it is confirmed that the proposed method was able to identify the target object with high accuracy.

Aging of skilled craftspeople and the successors lack have become serious issues for inheritance of traditional craft works. Solutions are demanded to preserve the craftspeople skills in a form to be passed on for lasting culture. In this study, we aim to extract skills of craftspeople as quantified knowledge through analysis of their works. We focus on knitting works, which require manual dexterity. we analyze the knitting skill by measuring hand motions and reaction forces of fingertips. In particular, we analyze effects of differences of hand motion on knitted patterns. Through measurement experiments, we consider hand motions of skilled work.

In recent years, labor shortages and the aging of the workforce have become problems. One solution to this problem is the introduction of humanoid robots to perform tasks on behalf of humans. Since humanoid robots have the same structure as humans, they can have a variety of motions by extracting features from human motions and estimating posture. In this study, we focus on the redundant parts of a humanoid robot and human that are not directly related to reaching the target positions of the hands and aim to analyze the stabilization of humanoid robot motions and the preparation posture for movements that require force. In the experiments, we focused on the putting and pulling movements of the shelf, which requires maintaining its balance and exerting force and confirmed the differences in the movements by changing the weight of the objects. We analyzed the movements by focusing on the flexion and extension of the torso and the opening of the elbows while holding the object close to the torso. From the results of the experiments, when the object’s weight was larger, the timing of grasping was earlier, the opening angle of the elbow was larger, and the torso was closer to the target object.

Upper limbs are often deeply involved in activities of daily living. Therefore, hemiplegic patients need to repeat rehabilitation. We have developed a training robot using Virtual Reality, Augmented Reality, and monitors with the aim of reducing the burden on the therapist and encouraging voluntary rehabilitation. We evaluated the safety and satisfaction of the rehabilitation. From the results, it was found that all training can be safely performed except for some people. We also found that training using Virtual Reality was more satisfying than training using Augmented Reality