Monitoring the walking ability in daily life is important for rehabilitation and keeping motivation for health. In this study, to develop a system to monitor the walking ability of cane users, we propose a method to detect the walking motion using an IMU sensor attached on the cane and to analyze the walking parameters for evaluating the walking ability. We detected the initial contact and terminal contact of the cane using jerk and angular jerk from the IMU sensor. We measured the walking of the cane users, and found variations of the cane angles of initial contact and terminal contact and time on ground of the cane might related to walking ability.

Decrease in balance control ability leads to the risk of falling among the elderly, and reactive postural control is important to prevent falling. In this paper, we developed a wearable suit for improving balance function with pneumatic gel muscles (PGMs). This suit can make perturbations to the upper body and aims to improve the function of reactive postural control through balance exercise. Small perturbations are generated by contraction force of PGMs attached the suit. Furthermore, the suit was designed to be lighter using DhaibaDAQ as the accelerometer and switch module. We described the system configuration, the torque of PGMs, and the control method.

This paper focused on cooperative skills in order to investigate physical human interaction key factors using a cooperative model. Human interaction is made up of both verbal and nonverbal communication. Especially, force communication has drawn a lot of attention as a means of nonverbal communication. Several studies focus on force communication, however, these studies have not considered human skills. In particular, cooperative skill is important for Human-Human interaction (HHI) or Human-Robot interaction (HRI). This paper proposes the neural networks that learn the CFO. CFO is estimated by comparison with learning the input command during the solo task using neural networks and the input command during the cooperative task. Also, this paper analyzes physical human interaction using the data which is compressed information by neural networks. In the experiment, the proposed neural networks can predict CFO precisely. In addition, the results show that followership on cooperative skills is induced by other players.

Understanding the reorganization of the central nervous system after stroke is an important endeavor in order to design new therapies in gait training for stroke patients. Current clinical evaluation scores and gait velocity are insufficient to describe the state of the nervous system, and one aspect where this is lacking is in the quantification of gait symmetry. We propose the use of muscle synergy analysis to quantify gait symmetry to solve this gap in knowledge. Results have revealed mechanisms of gait symmetry which could otherwise be difficult to observe with clinical scores. Furthermore, robotic gait training appears to provide an advantage over conventional gait training in restoring symmetrical neuromuscular control.

Human augmentation by using robotics has been exploring by various approaches such as Exoskeleton or Telerobotics. Furthermore, the research topic of Robotic Supernumerary Limbs (SuperLimbs) is recently studying as a novel approach for human augmentation. The research issues of SuperLimbs require perspectives and approaches in human-robot interaction different from previous research topics. In our research, we focus on the sensory feedback system for the operator to perceive the status of a Robotic SuperLimbs which attached to the body of themselves. Our approach is to use haptic feedback to provide proprioceptive information of the robot to the operator. In this paper, we consider the design factor for sensory feedback under the condition of SuperLimbs usage and describe a prototype of a wearable haptic device and system for haptic feedback. And then, we report the result of a preliminary experiment for system evaluation of the haptic device.

Large-scale distribution hubs handle a number of parcels having various physical properties (e.g. size, shape stiffness, etc.). However, traditional end-effectors are not versatile enough for robotic systems to grasp them. We have developed an original robot hand and a grasp planning algorism. The hand has a suction mechanism and three fingers actuated by two motors, and can grasp various objects by using two types of mechanisms individually or in combination. The effectiveness of the developed hand is evaluated by experiments using evaluation objects simulating field parcels.

This paper presents a two-fingered robotic hand using Twisted Round-belt Actuators(TbA). This mechanism generates contraction force by twisting the round belt, which enables to control micro force. Focusing on this feature, we develop a two-fingered robotic hand combining TbA with tendon-drive mechanism. First, the round belt is contracted by a DC motor, which generates contraction force. Second, a attached string is pulled by contraction force. Finally, each link bends as same as conventional tendon actuator. In addition, we develop a robotic palm which consists of springs and Polyacetal materials. Due to this, this robotic hand can absorb impact forces and protect grasped object. Based on the above, we show that this robotic hand can grasp soft object without damages.

Various multi-end effectors have been researched and developed, and it has become possible to grip various objects without requiring a complicated control. However, in tasks such as writing with a pen and assembling bolts and nuts, it is necessary not only to grip the object but also to replace the gripped object in hand (“ in-hand manipulation ”). The development of a multi-end effector with in-hand manipulation skill reinforced the versatility of the robot. However, realizing in-hand manipulation usually required a large amount of feedback from the sensor and complicated control. In this study, we developed a robot finger that realizes in-hand manipulation with easier control, especially for small objects such as bolts, nuts, and screws, by combining rotation with a finger that has a mechanism that uses the jamming transition phenomenon.

A conventional pre-grasping control using a photo-reflective proximity sensor keeps the distance to the object constant by keeping the total photo-current constant. However, since the total current depends on the reflectivity and shape of the object, calibration is required for each object. In this paper, we focus on co-point ( distance at which the sensor outputs of two photo-transistors are equal) as a physical index independent of the object, and propose a method to control the distance based on the magnitude relation of the sensor outputs. By this method, the distance to the object can be kept constant without calibration.In addition, by making the optical path limiting cover, the directivity of the LED of the proximity sensor is artificially changed, which enables the LED to play multiple roles in the pre-grasping control: distance correction and tilt correction.

The two-layer fluid finger with texture structure inside has been developed to realize the ability to grasp various objects with a small degree of freedom. In this study, the effect of texture structure in the two-layer fluid finger on resistible force is investigated. In order to evaluate the effect of texture structure on the grasping performance (resistible force), the fingers with some shape and direction of the texture pattern are made. From the results of the evaluation experiment of resistible force, we found that the resistible force increases in a right-angled triangle pattern (mounting direction 180ﾟ).

This paper proposes a high speed gripper capable of both closing and releasing. The gripper is composed of two fingers freely moved by springs, air cylinder connecting with wire for actuating the gripper. Through experiments, we confirmed its basic performance.

An end-effector with the ability to work in the unstructured environment is becoming essential in today’s scenario because of the increase in demand for robots to work in collaboration with humans. For such end-effectors, the sensors are essential. However, in conventional finger designs, it is not possible to cover the whole surface area with sensors. So, this current paper suggests a finger using a Remote Center of Motion mechanism to move the center of the joint rotation to the surface of a thick skin layer. Therefore a thick soft skin layer with a continuous surface can be realized. In addition, this skin layer also acts as a cushion for the objects, which allows it the ability to grasp fragile objects.

In order to automate the work in a garment factory, we developed a method of single sheet separation from piled fabrics by a roller hand mechanism with a rotating mechanism added to the tip of a robot hand. As a result, we succeeded in separating only single sheet of fabrics with a high probability. It is necessary to eliminate failures for practical use. However, it is difficult to eliminate failure completely due to the characteristics of fabrics. Therefore, we devised a method to determine the success of separation, created an actual machine, and evaluated the performance of actual machine by experimentation using an actual machine. As a result, we could not accurately measure the thickness of the fabric, but we found the possibility of realizing the judgment mechanism of separation success by improving the mechanism.

To play an active part in our daily life, a robot with a soft cover that can softly touch people has been researched and developed. Unlike industrial robots, these robots have our daily life as the main activity environment, so there are many opportunities to interact with humans. For this reason, research on ensuring human safety using tactile sense is important, and the ability to communicate through direct contact with humans is required.

On the other hand, robots that support human daily life often require tasks the ability to grasp objects. However, the force transmission via the soft cover involves nonlinearity, which makes it difficult to control the motion by force feedback.

In this research, we propose a method of correcting the output of the tactile sensor that can be realized and assumed the object gripping motion of the robot hand covered with the soft cover.

A design of driving system of an anthropomorphic finger mechanism is discussed. A novel motion taxonomy of the human finger is organized from the viewpoint of manipulation tasks. It matches with one’s experiences and intuition, and suggests that an appropriate use of passive joint mechanisms naturally emerges preferable contacts between an object and the fingertip. The designed mechanism is based on this and the anatomy, which successfully reproduced all motions in the taxonomy.

Currently, object identification is searched using machine learning or deep learning. However, it can be difficult to identify objects using only information obtained from images. Therefore, when identifying an object that can’t be determined only by visual information, the vibration transfer characteristics are measured using a piezo actuator and an acceleration sensor. Also measures the amount of deformation when a force is applied to an object. The purpose of research is to study a robot hand that has the function of active vibration sensing an object from these two types of information.

In this paper, two types of Universal Vacuum Chuck (in short, UVC) with a finger structure for grasping various objects are proposed. The UVC is a vacuum suction cup based on the jamming transition phenomenon, which is a combination of Universal Vacuum Gripper and robotic finger. One of UVC has a built-in soft finger, and the other type has an external hard finger. The adhesion and gripping performance of the two UVCs were evaluated. The soft finger type was suitable the low and wide uneven surface and was able to grip a small test piece. On the other hand the hard finger type was suitable high and narrow uneven surfaces and showed high gripping force.

Pizza has a number of toppings and human sprinkle them skillfully with their hands. When such a sprinkle operation is realized by a robot hand, it become possible to recognize the size, shape, and hardness of the object and drop it evenly, that is, by controlling the flow rate of the grain. Therefore, the purpose of this study is to recognize various information of objects and to evenly sprinkle using a hand. In this paper, we focused on the position, and made a hand to control the flow rate of the grain from the information.

Sprawling locomotion is a quadruped walking gait with lateral body bending used by salamanders, lizards, and crocodiles and so on. It presents an interesting example of how quadrupeds coordinate their legs and other body parts such as the trunk, head, and tail for adaptive locomotion, i.e., an interesting example of body–limb coordination. A better understanding of this body–limb coordination mechanism can provide informative insights into animal locomotion control and into the improvement of the legged robot mobility. Previously, we demonstrated that sensory feedback facilitates body–limb coordination in sprawling locomotion using a decentralized control with cross-coupled sensory feedback between the legs and the trunk. In this proceeding, we conducted the experimental validation of the proposed control using a developed quadruped robot. The resulting gait showed stable sprawling locomotion and presented the usefulness of sensory feedback for body–limb coordination, even in the real world.

Centipedes can move adaptively in unstructured environments by coordinating a large number of legs. Clarifying the underlying control mechanism for walking will help not only contribute to biology but also develop highly adaptive multi-legged robots. We previously observed the centipedes’ response to the removal of a part of the terrain during walking and reported that the adaptive locomotion could be generated by a control mechanism using ground reaction forces detected at the legs. In this paper, we additionally observed the response to the appearance of a part of the terrain and found that centipedes utilized the newly obtained scaffold for propulsion. Based on this finding, we proposed an inter-limb coordination mechanism for multi-legged robots that enables adaptation to irregular terrain, and as a first step, we demonstrated via simulation that the proposed model could reproduce a steady gait pattern of centipedes.

Quadrupeds exhibit versatile gait patterns in response to locomotion speed. In our previous study, we designed a simple leg-control rule that can exhibit gait transition from walk to gallop according to locomotion speed. However, high-speed locomotion was not realized due to lack of jumping motion. To address the problem, we designed a new leg-control model that realizes jump by modifying leg motion during the swing phase. As simulation results, we succeeded in realizing jump and widening the range of locomotion speed. In this study, we developed a one-legged hopping robot in order to verify the proposed model in the real world.

Snakes have various locomotion patterns and change them in response to the situations controlling their elongated body. The ultimate goal of this study is to develop a decentralized control scheme for snake-like robot that can reproduce the various locomotion patterns. For this purpose, in our previous study, we proposed several control schemes based on Tegotae, a concept describing how well a perceived reaction matches an expectation. Although these control schemes succeeded in reproducing some of the locomotion patterns, variety of the locomotion patterns were still limited. To address this issue, in this study, we propose a decentralized control scheme for snake-like robot that can reproduce the versatile locomotion patterns by integrating the essence of the control schemes we proposed previously.

Polychaetes have a number of body segments with a pair of parapodia. They locomote effectively by utilizing undulatory and peristaltic motions of the flexible body in response to the environment. Clarifying the control mechanism underlying such adaptive locomotion of polychaetes can contribute to developing multi-legged robots that can move effectively in various environments by the flexible coordination of many body parts. In this study, we observed the locomotion of polychaetes when a part of their bodies was placed in a narrow space, and found that the motions of each body segment adaptively changed in response to the situations. Based on this finding, we proposed a decentralized control scheme that can reproduce the adaptive switch between undulatory and peristaltic motions in polychaetes locomotion.

Insects adapt their locomotor behavior in response to changes in environment and context by altering both inter- and intra-leg coordination. To elucidate underlying mechanisms in the leg coordination, we apply an external and precise method of interference with neural motion control, based on the electrostimulation of leg muscles. Here, we propose the method, called “Motion Hacking”, based on engineering techniques. First, we investigated externally controlled joint torques induced by stimulating one out of three leg muscles (protractor, retractor, and levator) in the stick insect Carausius morosus. For a given parameter set of a burst of pulse-width modulated electrostimulation, we found a piecewise linear relationship between the burst duration and the generated joint torque. Linearity holds for a burst duration range between 100 and 500 ms, corresponding to the typical values of swing and stance durations of walking in stick insects. The result suggests that the extent and timing of movement generated by joint torques of a single leg can be controlled. This is a necessary prerequisite for hacking the motion of a leg via external muscle stimulation in free walking insects.

As one of the cooperation systems composed of a lot of robots, we focus on a system that cooperates with each other by using its own body like the army ant. In general, group cooperative behaviors require a lot of sensors and complicated networks. However, the whole system gets difficult as the number of robots increases. Instead of these complicated network-based approach, we aim to construct a system with the aid of the robot body, what we call the mechanical approach. In this paper, we conduct a climbing experiment to examine the validity of the mechanical approach.

In this paper, we present a pose graph map merging SLAM for multiple mobile robot system. The local maps from individual robots are merged in a global framework with little to no latency and the updated maps are sent to all the robots in the network to update the changes in the map. By merging the local maps from the robots, a global representation of the area can be used for different applications The process allows our system for data association and efficient loop closure even for large areas. By combining data from different sensors on the robot, the map merging process is initiated in the back-end while, the local mapping and corrections are done on the robot itself. RGB-D features from the robots are used to detect loop closures and areas in the map that have an overlap and these areas are used to further optimize the graph in the back-end.The proposed system is tested with real robots and the results are discussed.

This paper demonstrates “no-microcomputer-based” sheepdog-type robot navigation utilizing magnetic force. The sheepdog system is inspired by a herding event, in which a small number of sheepdogs controlled by a sheep herder manipulate up to one thousand sheep flock indirectly. This system is considered as interesting control system, because the sheepdogs who act as a small number of controllers can control multiple sheep that cannot be directly controlled by using their own maneuverability. In order to construct this system as simple as possible, we design agents interaction based on repulsion force of magnets. We produce pseudo magnetic monopoles composed of multiple neodymium magnets whose south poles are arranged so as to face the outside. Through the robot demonstration with twenty sheep robots and one sheepdog robot, we verify the proposed system.

Goal-conditioned reinforcement learning (GCRL) and graph search often have the same purpose. In other words, both aim to find a policy or a route to achieve a given goal. The former can accomplish complex tasks under uncertain environments, but tends to fail long-horizon binary reward tasks. On the other hand, the latter can calculate even a long-term route if environmental information such as a short-distance travel cost is given in advance. In this paper, we expand the Floyd-Warshall method, a graph search algorithm, for GCRL framework. We call it Extended Floyd-Warshall method. Moreover, we propose a new GCRL method that alternates between the conventional GCRL and the extended Floyd-Warshall method to handle long-distance binary reward tasks. We demonstrate that our method has outperformed the conventional GCRL methods by experiments in a grid maze environment.

We propose a learning algorithm for an autonomous robot to acquire the observation skill that is advantageous for achieving its task. We consider the robot has movement and observation agents separately; the observation agent learns a policy for providing the observation for the movement agent, which learns how to achieve tasks better. Each policy is trained separately, and the observation policy is updated by using the differential value function before and after the movement policy is learned by the observation given by itself. Experiments on 2D navigation tasks in simulation show that our algorithm is more successful than conventional methods for the situation in which agent’s view is narrow.

Imitation learning has been widely used for autonomous mobile robots. Imitation learning can be performed with deep neural network. Deep neural networks have feature extraction layers that grasp the features of the input and identification layers that identify a category of the input. This feature extraction layers can be reused to other tasks by a method called fine tuning. Fine tuning is expected to reduce the required training data. Fine tuning is used for various tasks. However, it has not been studied which neural network architectures and pre-training tasks are suitable for autonomous mobile robots. In this paper, we compare widely used neural network architectures pre-trained on both object recognition and image reconstruction tasks. We conduced experiments to compare the performance on mobile robot tasks among learner policies with several architectures and pre-training methods in a simulation environment.

Imitation learning is one of the techniques for autonomous control in robotics. This technique enables robot to autonomous control by human demonstrations. However, in a different environment from demonstrations, test of autonomous control based on learning result is often fail because domain that distribution of dataset between demonstration and test is difference. In domain adaptation, this approach transfers knowledge of an environment with sufficient domain to an environment with insufficient domain and it enables autonomous control even in an environment with a small amount of demonstration. We verify the domain adaptation to autonomous moving control by imitation learning and clarify the problem.

The purpose of this study is to take over the interaction between robot and human. This paper describes a hand recognition system. Training is performed by using the object recognition method YOLO. There are three versions of YOLO. The hand image dataset is split into three training datasets of 1,000, 3,000 and 10,000 images and is used for training three versions of YOLO. The evaluation is performed using the class-specific confidence score and the evaluation indexes. In the case of using 10,000 images, the average value of recognition confidence exceeded 90% in all methods. In the evaluation, YOLOv1 is considered effective when the number of classes to be recognized is small.

Autonomous mobile robot navigation system are fundamental and important in robot technology. In recent years, researches about autonomous navigation system based on reinforcement learning have increased. It is expected that autonomous navigation system based on deep reinforcement learning is achieved without grid map prepared in advance. However, in autonomous navigation based on deep reinforcement learning, the difficulties of action and reward design are mentioned. In this study, we aim at an automatic action policy acquisition system that designs them based on clustering of velocities and generative adversarial imitative learning from behaviors of expert robots. In this paper, we describe system abstract and develop system. Furthermore, we perform several experiments to examine the validity of the system.

The parallel link mechanism for robots is expected as a robot component for realizing such as high-speed motion, high-rigidity for a load, high-precision positioning, and so on. One of the parallel link mechanisms is the 3-DOF spherical parallel mechanisms. This is that the Base Plate is connected to the End plate with the multiple units of 3 joints and 3 links. We consider to use the 3-DOF spherical parallel mechanisms as a joint actuator, the shoulder or wrist of the robots, and so on. In this paper, we propose the inverse kinematics of the 3-DOF spherical parallel mechanisms using a spherical trigonometry and the forward kinematics using the coordinate transformation. Also we have made the trial products of the 3-DOF spherical parallel mechanisms for 3 arms and 4 arms.

Automated construction vehicles are required to compensate for the reduction of workers in construction industry. In one of construction works, earthmoving work, a dump truck and backhoe work in cooperation and some researcher have been developing the automation of a dump truck which could transport sediment. However, it is difficult for the automated dump truck to work with a backhoe which operated by a human. In leveling sediment, the dump truck must not get close to the backhoe. We build the motion model of the backhoe which is operated by a specific operator with the beta-process hidden Markov model (BP-HMM) and focus the pattern of primitive motions. The instance when the automated dump truck should move to the loading location could be predicted with the confidence by tracking each primitive motion in the pattern.

We add a reset algorithm using Hierarchical Dirichlet Process Hidden semi-Markov Model (HDP-HSMM) to Particle Filter on Episode (PFoE), which is a teach-and-replay method for an autonomous robot. HDP-HSMM is used for classification of sensor data on the memory sequence obtained in the teaching phase. The proposed algorithm for detects the discordance between the expected sequence of sensor readings and the actual one when the robot is replaying the taught motion. When the discordance becomes apparent, PFoE can retry the taught motion in at the middle of the motion by searching an appropriate recovery point. We show this method works on a small mobile robot.

In recent years, autonomous delivery system for UAV have been developed actively to improve logistical efficiency. Delivering to a small living space requires autonomous flight abilities such as obstacle avoidance and pinpoint landing. This study aims to develop a vision-based autonomous flight system for UAV that performs self localization and obstacle avoidance simultaneously based only on monocular camera images. Self localization is realized by estimating the position and the inclination of the AR-marker. Obstacle avoidance is realized by predicting the depth of a single image using Convolutional Neural Network(CNN). Verification experiments confirmed that proposed system has ability to fly and land UAV toward the AR-marker while avoiding obstacles even in a non-GPS environment.

Imitation learning is a popular method to obtain policies on autonomous robots given expert demonstrations. Recently, adversarial imitation learning methods, such as generative adversarial imitation learning (GAIL), have achieved great successes even on complex continuous control tasks. However, GAIL as well as its variants require a huge amount of environment interactions that often take impractically long time for training the robot. An intuitive way to reduce the number of interactions is initializing a policy by behavioral cloning (BC) before performing GAIL as pointed out in [1]. However, Sasaki et al reports that the BC initialization does not lead to reduce the number of interactions at all, rather significantly harms the imitation results. In this paper, we further analyze the BC initialization to figure out why the results are opposed to the intuition. Experimental results show that one of the cause of failure due to the BC initialization is that BC vanishes gradients of objective functions for the adversarial imitation learning algorithms, even though the objective differs from that of BC.

The development of reinforcement learning and deep neural networks allow us to train a decision-making system for robots by the end-to-end method, which directly leverages raw sensory inputs, and outputs an action. Designing a reward function that not only reflects the goal of the task but also facilitates the agent’s exploration, however, is tedious and challenging. This paper introduces a technique that allows agents to explore following the expert-designed state trajectory and take a balance between the creativity of agents and the rigid rules of the game shaped by prior knowledge. We investigate and evaluate our approach on a simple case and a complex robotic arm grasping-task. The results show that our method has a good application prospect in the sim2real field.

Synergy supplies a practical approach for expressing various postures of a multi-fingered hand. However, a conventional synergy defined for reproducing grasping postures cannot perform general-purpose tasks expected for a multi-fingered hand. Locking the position of particular fingers is essential for a multi-fingered hand to manipulate an object. When using conventional synergy based control to manipulate an object, which requires locking some fingers, the coordination of joints is heavily restricted, decreasing the dexterity of the hand. We propose a functionally divided manipulation synergy (FDMS) method, which provides a synergy-based control to achieves both dimensionality reduction and in-hand manipulation. In FDMS, first, we define the function of each finger of the hand as either “manipulation” or “fixed.” Then, we apply synergy control only to the fingers having the manipulation function, so that dexterous manipulations can be realized with few control inputs. The effectiveness of our proposed approach is experimentally verified.

The automations of a convenience store such as Amazon Go is one of the hot topics. In the future, it is expected that the refilling task of the convenience store items are automated by robots. The items in a convenience store are regularly arranged on the shelves with the same items to improve storage space saving and appearance. It is very important to recognize the object arrangement pattern and detect the items to realize the automation of a convenience store tasks. This paper reports the recognition of the object arrangement pattern and detect the individual items using machine learning method.

Dual-arm robots are capable of handling objects as well as humans. Such robots have potential for supporting human life. For example, in order for a robot to pick up an object, it is necessary to autonomously generate the reaching motion toward the object. For this issue, we have thus far presented an End-to-End motion planner based on convolutional neural network, CNN. However, since a single object was assumed, the robot based on this motion planner fails to generate the reaching motion for multiple objects. For this challenge, we focus on an image segmentation technique, which is so-called semantic segmentation. In the learning phase based on CNN, segmented images of the reaching target are used as the input. Through the experiments, we show that the robot is able to generate the reaching motions toward multiple objects.

There are many studies on multi-arm cooperation. They usually require expensive force sensors and complicated force control. We focus on dual-arm cooperation using only position control. However, dual-arm cooperation without force control may cause excessive internal forces due to position error between two arms. To solve this problem, we propose the use of caging to constrain objects. Caging is a method of object constraining based on geometric information. It allows an object to move within the hands of arms. Caging enables arms to avoid excessive internal forces in position-controlled manipulation. This caging-based dual-arm cooperation was carried out successfully in some experiments.

In recent years, reductions in the labor force due to aging have become a problem in developed countries. Therefore, automation is advanced in multi-product variable production which has been conventionally done manually. However, it is still done manually in terms of tasks requiring hand dexterity such as using tools, and handling of flexible items such as foods. Therefore, in this research, in order to realize the automation of the manual work, we aim at the construction of an object grasping system that capable of dexterous, soft and robust grasping. In this paper, we introduced a stability index based on potential energy and constructed the gripping simulation.

Redundant manipulator is a robot which has more than seven joints. This robot can achieve the target trajectory of the hand while changing its shape according to the work purpose and the surrounding environment. With this feature, this robot can perform wraparound and crouching motions by using elbow motion, realizing space saving on the production line. Direct teaching is a teaching method in which a human teaches a motion by moving a joint of the robot directly. It is intuitive and can be taught easily by non-experts. However, it takes long time because each joint needs to be moved, and it is difficult to move the robot with human power when using a reducer with a large reduction ratio. Therefore, we propose a direct teaching method with force control of a 7-axis robot arm that is a redundant manipulator in consideration of obstacle avoidance by using the elbow motion.

Robots are required to adapt flexibly to environmental changes in order to work in various environments. One of the effective ways to adapt flexibly to environmental changes is to estimate the impedance of the environments. The parameters of the impedance control is designed based on the impedance of the environments. The purpose of this study is to estimate the impedance of a high stiffness environments from the position and force information of the robot’s hand. In proposed method, exceptional handling of numerical calculation errors enables stable stiffness estimation. The estimation of the stiffness in one axis is shown by verification through experiments.

Visual servoing is capable of positioning robots based on images captured by cameras. To calculate the command value for robots, hand-designed image features and extraction of image features are required. Positioning accuracy is significantly influenced by the selection of the image features. In this study, we focus on the ability of convolutional neural networks (CNN) to extract features from images and output the angular velocity to control a manipulator. We propose a visual servoing technique based on CNN enabling the precise positioning of a texture less object grasped by a parallel gripper. The positioning can be achieved even the grasping position is different from the position when the target image was captured. The positioning accuracy of the proposed method is verified based on numerical simulation. We confirmed that the proposed visual servoing technique can position an object precisely.

This paper develops a mechanical tool for 2-finger parallel robotic grippers. It specially studies a mechanism that converts the gripping motion of grippers into continuous rotation so as to realize tasks like fastening bolts. The developed tool can output the clockwise rotation at the front end and the anti-clockwise rotation at the back end. Exchangeable rotate tooltips can be installed on both the ends. Robots may autonomously decide the tooltip and rotating direction, install the tooltip onto the tool, and adjust the pose of the tool to complete given fastening tasks. The essential structure of the tool is based on a Scissor-Like Elements (SLE) and a double-ratchet mechanism. They convert repeated linear motion into continuous rotating motion. The purely mechanical design allows robots to use the tool without any peripherals and power supply. This paper describes the detailed design, the structure optimization, and implements experiments like fastening bolts to analyze its performance and demonstrate the advantages.

This paper describes pitching a ball using parallel elastic joints arm. To pitch a high-speed ball, it is necessary to obtain high kinetic energy. By resonating with the spring-mass system, the arm can obtain high kinetic energy and the arm achieve high speed. On the other hand, when performing high-speed motion using a robot arm, the motion is restricted according to movable ranges. By considering the relationship between the joint limits and possible maximal speed and energy, high-speed motion can be achieved without exceeding movable ranges. We propose a method that is based on mode to obtain high speed with joint limits.

There are many tasks that require a constant distance from the object, such as scanning with a hand-held metal detector or spray painting. However, in the teleoperation of the robot, it is difficult to understand the distance. In this paper, in order to solve this problem, a control method is proposed that compensates the distance between the object and the end effector using three-dimensional point cloud data. And, the experiment which moved the end effector in front of the object with the roundness while keeping the constant distance was carried out. As a result, it was shown the effectiveness of the proposed method by comparing the trajectory with and without the control.