This paper reviews a dynamics discretization method called spatially quantized dynamics (SQD) and shows that time scale transformation yields the SQD. Time scale transformation also derives continuous spatial dynamics representations, which are equivalent to the original dynamics. By introducing a path parameter and path frames, the SQD are expanded into the 3-dimensional Euclidean space. An efficient method to calculate actual time steps evolving system state is proposed. As a result, trajectory optimization using the proposed SQD enables a kinematic walking pattern to satisfy dynamic constraints, and how to generate a biped gait with stretched knees along an arc on a horizontal plane is presented.

Deformable objects including foldable stands, binders and tripods are necessary in our lives. Therefore, if humanoid robots, which are imitating human figures, can also use such objects, the range of tasks that robots can do will expand. Then a method to sample states of deformation is effective for motion planning including such objects. However, we have to consider how to sample them and how to express a transition between them. In this paper, we propose a method to acquire a model considering the transition. The method involves teaching of geometric shapes and grasping positions of especially important deformation states. We also propose a motion planning method that can consider the deformation of the object using the obtained model.

The purpose of this study is to make life-sized humanoid robots acquire tool manipulation skills that require complicated force adjustment. The difficulty in acquisition of tool manipulation skills comes from the hardship in physical modeling. Recent research has revealed that deep reinforcement learning (DRL), a model-free approach, performs superior in such tasks. However, DRL in general has a drawback in sample efficiency, and this becomes critical in robot learning especially in life-sized humanoid robots. In this study, we propose an integrated system incorporating DRL method and active learning. Our method also leverages a variety of previous studies on life-sized humanoid robots to overcome the sample efficiency issue. We demonstrated the effectiveness of our proposed system through a hacksaw skill acquisition and a Japanese planer (Kanna) skill acquisition by a life-sized humanoid robot.

This paper discusses a compliance distribution optimization that satisfies positive definiteness. The compliance distribution optimization is a method to realize different types of compliance in the task space by optimizing compliance in the joint space, minimizing the distance from its referential value. To evaluate the distance between matrices, we use Riemannian geodesic distance that allows us to consider the geometric structure of positive definite symmetric matrix. In order to satisfy constraint on the task-space compliance and positive definiteness of compliance, we propose using the null-space of Jacobian matrix that maps joint space to task space.

When the environment or the work changes while the robot substitute humans’ works, the robot can be adapted to various situations by changing to a form suitable for such as a mobile-base or humanoid depending on the situation. Therefore, in this research, by constructing a four-wheeled humanoid that has driving wheels, trailing wheels, and multiple sensors, capable of transitioning between multiple forms and returning from falling to continue operation, we examine the construction method of the robots that have continuity of movement and can take various forms and the form transition method. Experimental results show that, it is effective (i) changing the support area with driving wheels and trailing wheels actively, (ii) light and flexible impact resistance of the body, and (iii) detecting the state of the actual robot using multiple sensors equipped on the whole body.

In this paper, we focus on humanoid robots rough terrain traversal. In order for humanoid robots to cross over uneven ground, we propose a humanoid robot mannequin device with actuators. The device has the same number of joints as the controlled humanoid robot and the operator of the robot can switch which limbs the operator would like to control. With the mannequin device, we did an uneven ground traversing experiment with real-time operation.

In recent years, humanoids are expected to be used in real environments. However, humanoids are unstable and it is difficult to completely prevent falls. For this reason, it is necessary to take appropriate measures against falls that cannot be prevented. However, life-sized humanoids are large in mass and contain many fragile mechanisms, therefore they are vulnerable to impact. In this paper, we propose an exterior that uses an air damper shock absorbing exteriors that hardly restrict robot’s movement, is effective even when the robot system is stopped, and can be used for repeated falls. Furthermore, We propose a method for automatically generating Ukemi motion that make effective use of exteriors. The proposed method was applied to a life-size humanoid RHP3 and a fall experiment on a hard floor surface showed that the maximum acceleration at the time of collision could be reduced to about 40 [G].

The purpose of this study was to compare the standing up movement of hospitalized rehabilitation patients with healthy young people and to investigate the characteristics of exercise. In addition, we surveyed the current status of the standing up support system that was developed, and examined the specifications of the support system required for rehabilitation patients. In the standing motion of rehabilitation patients, there was a remarkable left and right difference in the change of each joint angle during the standing motion compared with the healthy person. Also, when observing the movement of the center of gravity from the front, the displacement of the center of gravity in the left and right direction was large, and the asymmetry of the movement was remarkable.

Recently exploring for a possibilities different from conventional robots, various type soft actuators and its applications have studied. Especially, pneumatic soft actuators have the advantage in lightweight and powerful. The features of this idea are simplified systems, almost all interfaces can be configured as a flexible structure. In particular, pneumatically driven soft actuators have been studied. Since these are driven using a limited pneumatic path, the pneumatic path usually increases in proportion to the degree of freedom. In this study, we develop to generate multiple deformations in a foam laminated foam based soft actuator using one air pressure path by using the difference in compression characteristics of foams, especially Open cell foam and Full open cell foam against pressure.

Recently, unmanned and automated systems using robots have been developed and used in various fields. These robots can be returned for reuse, or discarding after the intended work. Since recyclability of the materials used for robots is desirable in some applications, various biodegradable actuators that produce motion are the focus of current research and development. In this study, we develop a biodegradable soft actuator with a 0.1-m class size, and analyze its disassembly performance. Specifically, we focus on bread as a biodegradable material that can be easily formed into various shapes. We propose a bread actuator driven by using the sponge foam structure of bread. This paper reports on the basic concept of biodegradation in the food chain, which becomes animal food after use, as well as its operation principle and feasibility.

The authors have developed A variable viscoelastic joint that mimics the principle of human joint drive and uses a pneumatic artificial muscle with an antagonistic arrangement and a magnetic viscous fluid brake. In this paper, the versatility of this joint was improved by modularizing this joint. The variable viscoelastic joint module is the world's first pneumatic drive module with a variable viscoelastic characteristic that includes a power source inside the device. We believe that the development of this device will make it easier to manufacture robots that are lighter, safer and easier to maintain than robots that use motors and reduction gears, and will operate at lower cost.

The research proposes a particular tensegrity topology for a humanoid hand in which the motion of the hand is actuated using thin McKibben artificial muscles. The specific actuation pattern for this tensegrity hand with multiple degrees of freedom, which includes the location of the artificial muscles, is presented. The humanoid tensegrity hand is able to resist impact such as compression. After resisting impact, the tensegrity hand returns to its original shape and works normally. The related experiments were conducted to evaluate the characteristic of impact absorbing. The implementation of the tensegrity hand shows that this specific topology design has potential to develop soft robotic hands.

In the inspection of sewer pipes, it is necessary to identify the damaged part, and a pipe diagram is indispensable, but it is often dissipated and many pipe maps do not exist. Therefore, it is required to create pipeline diagrams. Therefore, the authors used an IMU sensor that measures three-axis acceleration and three-axis angular velocity, and corrected the cumulative error that occurred at that time by using the traveling characteristics of the earthworm's peristaltic motion, which is a periodic motion of progression and stopping. We tried to measure the pipe length and the pipe shape. In this report, we conducted a basic evaluation experiment to measure the pipe length with high accuracy, and investigated the effectiveness of this method.

In the recent years, aged pipes are increasing, and many road collapse accidents occur. Therefore, it is necessary to inspect the condition of the pipe to efficiently determine whether a replacement or repair is necessary. However, pressure pipes that are capable of coping with complex terrains, the pipelines are complicated and are difficult to inspect with the existing inspection equipment. Therefore, based on the earthworm's peristaltic movement and artificial muscles, the authors developed a peristaltic robot in a previous research for inspecting the inside of a pressure-feeding pipe. However, owing to the unevenness its joint, the robot was unable to maneuver through the corners of a curved pipe. Therefore, in this study, we developed a high rigid joint attachment that enables the robot to pass through a curved pipe, and evaluated its effectiveness through traveling experiments by using a curved pipe.

Using a Actor-Critic algorithm, a stable motion pattern is determined for a multi-segment robot composed of a linear actuator with a longitudinal muscle function and three servomotors with an annular muscle function. The operation was confirmed by adapting to the actual machine.

High viscosity fluids, solid-liquid mixed fluids and powders are used in various fields such as industries, foods, medicine, etc. as products and materials. Existing methods such as screws may be unusable due to the shear forces that arise. Therefore, the authors developed a peristaltic motion type mixing conveyor using artificial muscle. Larger the device is required for practical use, but a large space is required for driving it. Reduction of expansion diameter for practical use will contribute to improvement of mixing conveyance efficiency and cost control. Therefore, by attached ring to artificial muscle, we aim to reduce the expansion diameter required for operation.

The musculoskeletal system of any animal changes by growth, aging, and injuries. Motor learning is needed to adaptively compensate for these changes. The cerebellum plays a crucial role in motor learning. To date, several neuromorphic cerebellums (artificial cerebellum) have been developed on supercomputers and PCs. In this study, we implemented an artificial cerebellum on a field programmable gate array, and applied it to real-world machine control. As a simple example, the artificial cerebellum was employed to control a DC motor whose load was fluctuated during a constant velocity driving task. It was demonstrated that the artificial cerebellum adaptively compensated for the load change and reduced control errors.

For designing a mobile robot, we focus on peristaltic movement, which is motion pattern of earthworm. To realize flexible body, a peristaltic robot using a servomotor and a timing belt is proposed. The prototype of the peristaltic robot is designed and constructed. The evaluation of the motion is demonstrated through experiments.

A steep cliff wall is one of the extreme terrains and inaccessible for conventional mobile robots. To cope with such a difficult terrain, multi-limbed free-climbing robots have been studied. However, development of an adaptive robot gripper to terrain roughness with mechanical efficiency still remain to be realized. In related works, although a spine-type multi-fingered gripper has been validated for robotic free-climbing, the previous grippers are oversized, hyper-redundant, and non-durable systems. This study proposes a tracing grip such that minimum fingers can achieve more adaptive spine-type gripping to the terrain surface. The tracing motion enables to guide the spine to a suitable point on the rough surface to be gripped or hooked. The developed gripper was validated through the gripping experiments using simulated terrains.The gripping performance of the gripper is quantitatively evaluated, and the effectiveness of the tracing grip is also shown.

Nowadays, there are a lot of children with congenital forearm deficiency. Therefore development of prosthetic hand for children is important. It is necessary for the child prosthesis to be small, close to the appearance of human hand. However, many of the pediatric prosthesis currently being developed do not have the interphalangeal joints (DIP joints), Proximal interphalangeal joints (PIP joints), metacarpophalangeal joints (MP joints), and interphalangeal joints (IP joints). In this study, we have focused on underactuated mechanism with gear trains. This mechanism is useful to reduce the number of actuators, and several joints move in conjunction to grasp.

We propose a new arrangement of LEDs and detector and a digital filter for our proximity sensor to realize multiple points sensing of an object surface within 1ms. The accuracy and calculation time were verified by simulations. From these results, the proposed method enables a higher density arrangement than the conventional arrangement. The processing circuit can be downsized by replacing a conventional analog synchronous detection circuit with the proposed digital filter.

In this research, a gluing-less shell gripper is proposed for packaging multiple cucumber simultaneously. The proposed shell gripper consists of multiple thin flat fingers which were constructed by a thin shell and a soft chamber. The flat finger can be pneumatically inflated for the purpose of grasping objects. The flat finger was fabricated by only one-time casting process without demolding and gluing operations. This makes the gripper easy to be fabricated and having a thin geometry. The durability and pressure resistance were tested, and packaging experiments were performed to demonstrate the effectiveness of the proposed shell gripper.

In this paper, the powder of Universal Vacuum Gripper (in short, UVG) is included in stocking and filled into UVG. Filling the stockings prevents the powder from being biased vertically. In addition, by separating the powder in the stocking, the adhesion performance is improved. We evaluated the effects of division members for adhesion performance.

Universal Vacuum Hand (in short UVH) based on jamming transition of conductive granular material is expected as one of the universal robotic hand with tactile sensing. The principle of UVH's tactile sensing is electrical characteristics of the conductive powder inside the gripper which changes its electrical resistance due to the change in the density of the powder by outer force. In this research, we revised the previous structure of UVH to improve its sensitivity by separating the conductive powder by the membrane and evaluate the effectiveness and performance of the sensor-integrated gripper that was developed.

We have proposed a jamming membrane mechanism and a one-dimensional jamming mechanism as variable stiffness mechanism. These can be used as grippers that can firmly grasp complex shape objects. However, it is difficult for the jamming membrane gripper to grasp sharp objects, and for the one-dimensional jamming mechanism gripper to grasp fine objects. To solve these problems, we proposed a covered jamming mechanism. This mechanism encloses a striatum composed of beads in a rubber membrane. First, the basic principle of the devised mechanism was explained. Next, a basic experiment for holding an object were conducted using a covered jamming mechanism gripper based on the principle.

We examine transient response and asymptotic stability of pre-grasp control using a proximity sensor. In the past, we proposed simultaneously control of fingertip position and posture by combination of the proximity sensor feedback and an integral control. However, the analysis of transient response and the stability were insufficient. In this paper, we analyze the response and the stability after deriving the control equations. As a result of analysis, we confirmed that the response could be adjusted with one gain. In addition, we confirmed that the maximum gain could be determined in advance by using a stability discrimination.

Contraction Universal Gripper based on porous granular material (CUG) is one of the universal gripper for multiple object handling. Compared with Universal Jamming Gripper with hard powder, CUG has an advantage on the fragile object because of the softness of the porous granular material. However the porous materials takes time to be expand to normal state, thus the repetability performance was the problematic in case of the robotic applications.In this study, we focused on the intake and outlet shapes of CUG. Specifically, CUG has an exhaust and intake ports with angles and shapes. Repetability performance of CUG was clarifiedby measuring the gripping force (jamming gripping force) due to the jamming transition. As a result, the repetability performance was improved by giving the intake and exhaust ports an angle.

In recent years, the labor shortage in low-productivity industrial fields is escalating. In particular, handling products with various shapes and characteristics currently relies on human labor as can be seen in sorting work at convenience stores. In this study, we propose an end-effector that can lift variously shaped objects. The parallelogram link structure achieves weight saving and long stroke. The rotating fin structure enables packing work even if the gap between products is small.

Most of the robot hands difficult to deal with complex objects and control that many actuators due to complex mechanism. For this reason, inferior drived robot hand focused that a small number of actuators and high versatility to accommodate various shapes. On the other hand, robot hands using the jamming transition phenomenon are being developed. However, these are difficult to model and analyze the basic phenomenon, and aren’t suitable for precise control. We proposed a robot hand using an iris mechanism that made it possible to hold small and flat objects to large objects by making the blade shape conical and using a fin grip at the blade tip. We verified its effectiveness by experiment.

To automate the process of serving lunch boxes in the food industry, we will develop a gripper that can stably hold foods such as chopped green onions and granular corn. A four-finger wrapping gripper encloses food material almost without gaps. Due to the constant volume inside the fingers, the gripper can pick up food material stably. Four fingers made of silicone rubber are fabricated through casting. We evaluated the performance by comparing the amount gripped by the gripper with the amount served by human hands. In addition, the tendency of the grip amount with the different internal volumes was verified.

In this paper, we propose a multi-finger robot hand with iris mechanism. This new robot hand is based on the iris robotic mechanism. Construction of the new robot that blades of iris robot hand which attached each finger are denuded outside of mechanism enables gripping the long target lying on its side, tiny target such as M3x5 bolt without height adjustment base. Then new robot hand can grip more kind target than a conventional iris mechanism robot hand.

In this paper, we have developed a grasp assist system that can handle multiple grasp motions and suppress the effects of disturbances. In the system, the master’s grasp motions are mapped into the latent space using GPLVM, and are converted to the slave’s motions using GMR.In the experiment, it is verified that the proposed dimensionally reduction method is valid for the grasp motions, and, a three-fingered robot hand can be remotely controlled by the proposed grasp assist system.

An reinforcement learning, in which a robot acquires control law automatically due to interaction with an environment, is attracting attention. In this research, we had been investigated the acquisition of the control law for vehicle speed control in a field robot for forestry support. The final goal is to obtain the optimal control law for following a given refference velocity without modeling of the complicated terrain contained the work space and the complicated power transmission system of the robot. In this paper, we explain the acquisition of control law based on Deep Q-Network (DQN) and its design using actual robot. Finally, the effectiveness of DQN in speed control of field robots is described by comparing basic PD control and DQN control using actual robot and showing the significance of it.

Recently, various algorithms of deep reinforcement learning have been proposed. These are promising approaches to be applied to autonomous mobile robot. We have successfully implemented an autonomous mobile robot system to be navigated in real worlds using DDQN algorithm. Although DDQN provides only discrete actions, continuous actions are actually expected for robot navigation systems. In this paper, performances of action selections and autonomous navigation to destinations are compared among deep reinforcement learning algorithms. Especially, mobile robot systems by DDQN and DDPG are implemented in a simulator and evaluated.

This paper describes an automatic transfer rate adjustment method for transfer learning in reinforcement learning. Recently, learning robot systems such as reinforcement learning and transfer learning have been discussed. However, learning robots using transfer learning have a state of over-fitting that cannot behave correctly in an environment different from the environment in which they have been trained. The transfer rate and the transfer surface are proposed to avoid this over-fitting. The optimal transfer rate can be automatically estimated by using the transfer surface and the optimal transfer rate estimation method, however there is a problem that it is not possible to distinguish between two states, “Without solution” and “Solution with a width”. In this paper, we aim to prevent over-fitting by using a transfer surface, estimate the effective transfer rate in consideration of the problems in existing research, and discuss a method for automatically adjusting the transfer rate.

This paper is discribes the cognitive economics of the transfer reinforcement learning method using policy selection based on spreading activation model. Proposed policy selection model has some problems: calculation cost is increased by number of reusable learned policies. The computational processing cost increases because of aiming for higher cognitive functions in intelligent robots. As a basic consideration, parallelizing of policy selection method is proposed to increase the speed of processing. In this paper, effectiveness of parallelized policy selection method is confirmed using computer simulation and demonstrated using actual mobile robot.

Imitation is one of the important factors to acquire skills. Human generally acquire skills by mastering what he imitated from others. In this study, we verified whether it is possible for a human-like agent to learn things by performing imitation and mastery consistently like humans, and to be able to perform predetermined tasks. The task we focused here is the generation of walking motion for humanoid agents. We used Generative Adversarial Imitation Learning for imitation learning, and Trust Region Policy Optimization for reinforcement learning. We verified that it is able to generate walking motion for humanoid agent by combining these learning techniques.

The objective of this paper is the proposal of an active learning algorithm for robust manipulation. To this end, we introduce the concept of the feasible region and propose the size of the feasible region as a robustness measure of a trajectory. Roughly speaking, the feasible region is a region of acceptable perception error when a certain trajectory is executed. We then propose a method to obtain a robust trajectory by maximization of the robustness measure. Maximization and estimation of the feasible region size are performed simultaneously by the active learning of a real robot. The great thing of our approach is that a feasible region reflects physical interaction between the gripper and the object while the corresponding trajectory is executed. We applied our method to the door opening task of microwave-oven. After active learning, the robot acquired a robust manipulation skill which takes advantage of physical interaction with the microwave-oven.

Behavioral cloning, which is one of the imitation learning methods, enables a robot to imitate an expert’s policy from the expert’s state and action demonstrations. In that case, the robot does not need to interact with environment, thereby preventing robot failure. However, in general, it is difficult to obtain expert action information. Although behavioral cloning from observation allows the robot to learn the policy without that, it requires a few interactions with the environment to infer expert action, which leaves the risk of robot failures. Detecting faced situations are safe or dangerous is an effective way to prevent such dangerous interactions. Suppose that the expert’s demonstrations only visited the safe states, this paper proposes a new outlier detector using variational autoencoder learned by the expert’s data. It can easily find unexperienced and dangerous scenes since all the data used for learning are mapped to limited space. The proposed method improved the policy performance in simulations with the limited number of robot failures.

The authors have been studying on application of deep reinforcement learning methods to an autonomous mobile robot in target reaching tasks avoiding obstacles under uncertainties. In this report, randomly moving obstacles are newly introduced and effects of state selections on learning performance of Double DQN are examined. Learning simulations are performed using a robot model TurtleBot3 burger with an onboard 2D-LiDAR using a revised Gazebo simulator. As a result, it is confirmed that the number of success in target reaching avoiding obstacles during the learning process increases when the distance and the direction of the nearest obstacle from the robot are included in the states selection.

This paper addresses how to control a robot arm with variable stiffness actuators (VSAs) using deep reinforcement learning. Each VSA has two manipulated variables, target position and stiffness, thereby making learning speed inefficient due to increase of action space. To avoid the increase of action space including deviations of actuators’ stiffness, two policies for position control and stiffness control are explicitly distinguished, instead of end-to-end learning. In a ball throwing motion, the robot therefore learns how to swing its end effector firstly, and later, how to adjust stiffness to accelerate swinging motion. As a result, the robot in a dynamical simulator succeeded in acquiring the ball throwing motion without suffering from huge action space on end-to-end learning.

Aquariums have the task of water clarity testing for regular water exchange. Water clarity can indeed be detected by chemical reagents or special equipment, while the method is very tedious and complicated. Recently, artificial intelligence has been put to practical use in society, and it can contribute to the efficiency of water exchange using Deep Learning algorithms. In this research, we confirmed that water transparency can be determined using Convolutional Neural Networks, and that the water transparency determination achieved more correct answers better than the human eye. It was shown that a majority decision could improve the accuracy compared to classifying a single image.

In this paper, we propose a method to estimate the pose of an object grasped by a robot hand. Focusing on object manipulation tasks, we aim to prevent the manipulation tasks from failing because of the wrong pose of the object grasped by robot hand. To estimate the pose of an object, we compare the simulation and the real-world experimental data obtained by using RGB hand cameras attached at the wrist and a vision tactile sensor (using RGB camera) attached at the fingertip. We were able to estimate the object pose in the robot hand with accuracy of 55%.

We explore designing a guideline of robotic hands that can perform a range of manipulation tasks with minimum mechanical complexities. Referring to the analysis of human hand movement, we analyze finger-independent movement requirements for each manipulation task, and robotic hands that are capable of those movements. This paper reports a preliminary version of these analyses and its verification with two robotic hands that have different mechanisms.

In this study, we propose a method of modeling unknown objects using a dual-arm robot with a depth sensor installed on the head. In the proposed method, the posture of an object is changed by dual-arm manipulation , to obtain the shape information on the occluded part. First, the posture of the object is changed by pushing to obtain the point cloud of the occluded part with the depth sensor. Then the graspable plane is detected and the object is grasped to obtain the point cloud including the bottom surface of the object. Re-gripping is performed until the object can be placed such that the bottom faces upward. We further acquire the point cloud on the bottom and integrate the obtained point cloud. We evaluate the completeness of the model and the versatility of the operation by experiments.

This paper proposes a method for packing various objects into a container by a robot. Assuming multiple kinds of grippers to pick an object, we place each object stably into a container with checking the IK and the collision with neighboring objects. Suboptimal packing is realized by using the BL method along with the IK and collision checking. We finally fill the extra space with cushioning material. The effectiveness of our proposed method is confirmed by simulation results.

Robotic hands with soft surface can grasp objects stably, but cannot manipulate easily. To address this issue, we developed friction change mechanism called CAVS, which has high and low friction modes. These friction modes change reactively to applied load. This paper explores an internal sensor to estimate the friction mode of CAVS by detecting the deformation of its surface. In this research, we use a small camera to observe deformation of internal structure of the CAVS. The camera installed on CAVS was implemented to the fingers of parallel gripper and its performance was evaluated by picking-releasing task.

To automate bin-picking and kitting applications for texture-less objects with utilizing general-purpose hands, changing the grasping pose is required when the grasping pose is not suitable for kitting. We present a regrasping method to change the grasping pose with general-purpose hands mounted on a dual-arm robot. The success rates of the regrasping with general-purpose hands becomes lower due to a grasping error that occurs on the picking process. The proposed scheme corrects the grasping error before the regrasping by active visual servoing. Active visual servoing performs a visual servoing by projecting a pattern to the object. Active visual servoing does not require feature extraction of target parts; thus, positioning accuracy does not depend on the richness of texture. Experimental results show that the proposed method can achieve high success rates for kitting three target objects, compared to the conventional image-based visual servoing method.

This paper presents a nonprehensile manipulation method using a vibrating plate. Based on friction property difference, three-DoF (degree of freedom) motion of an object on the plate is generated using a single actuator. First, we show the velocity of the object with respect to the input frequency of the plate vibration. Its characteristics vary depending on the vibrational orbit and the surface friction property of the plate. Then, we discuss velocity vector fields on the plate induced by the combination of non-uniform friction property distribution and the input frequency. For the object placed on the boundary of two areas with different friction properties, the three-DoF motion of the object can be generated. Using a prototype, finally we show experimental results for confirming the feasibility of the proposed method.

In recent years, the shortage of labor in restaurants has become a serious problem due to the decrease in the working population and the serve working environment. One of the solutions to this problem is to introduce a robot system for automated dishwashing process. Therefore, robot hands for the dish washing process have been developed. However, these conventional robotic hands were designed for relatively lightweight dishes, and the design of the robotic hand to heavy iron plates is still a challenging problem. In this study, we develop a robot hand to be able to hold an iron plate.

First, we analyze the work process of the washing process for the iron plates to determine design parameters. Next, we develop the robot hand based on the parameter, and do the experiment to show the effectiveness of the hand.