In this study, we propose a shared control method for immersive teleoperation systems. The shared control method is intended to assist human operators in performing tasks that require dexterous manipulation, such as pouring a drink from a bottle into a glass in a remote environment. In this shared control scheme, the position and center of rotation of the bottle is adjusted and constrained based on the detected position of the glass. It also detects the height of the plastic ball in real time and assists the operator to fill the ball with the targeted amount. Experiments have confirmed that the system is able to accomplish the task as intended by the operator, even with irregular time delays.

This manuscript focuses on modeling of two-dimensional food shapes. Foods have much variation in their shapes. This paper presents a method to describe two-dimensional food shapes with variations based on approximated shapes and functional expansions.

In this paper, we propose a method of manipulation planning for cable wiring. The method enables to take into account the deformation of cables while connector incorporation process. Using a simulation that predicts the shape of a cable based on the posture of both ends of the cable, we generate a connector manipulation path that avoids contact between the cable and obstacles. We conducted simulation and confirmed that the proposed method can be used for inserting a connector into a socket installed on a circuit board.

Robot manipulation of flexible object is important and practical field of research. In this paper, we focused on rope turning motion as an example of manipulation of flexible object. We proposed a method to generate hand trajectory for rope turning based on length of a rope, and we realized stable rope turning motion by dual arm robot by our proposed method.

In recent years, the number of pipes that have exceeded their service life is increasing. Therefore, earthworm-type robots have been developed to perform regularly inspections of sewage pipes. However, inspection methods have not yet been established. This paper proposes a method for anomaly detection from images in pipes using Generative Adversarial Network (GAN). A model that combines f-AnoGAN and Lightweight GAN is used to detect anomalies by taking the difference between input images and generated images. Subtraction images is used to estimate the location of anomalies. Experiments were conducted using actual images of cast iron pipes to confirm the effectiveness of the proposed method.

This paper proposes a robotic manipulation method to reproduce the process of human’s food bolus formation. Decomposing and simplifying complicated mastication behavior of human, we define three primitives for robot manipulation: crushing, mixing, and gathering. We design the structure and motion of robot to execute the primitives, and develop a prototype. Executing these primitives sequentially, we try to reproduce the process of human food bolus formation. In experiment, human and robot masticate food samples and their food bolus images are acquired. By analyzing the images with convolutional neural network, we evaluate whether the robot can reproduce the human’s food bolus formation.

To shorten the work time (cycle time) of the conventional pick-and-place operation by robots, we propose a method for a robot to achieve pick-and-toss operation through learning. In this paper, for simplicity, the optimal hand opening width and arm joint velocity are obtained by learning for a tossing motion with a reduced amount of rotation to the target box. In particular, by considering whether or not the projectile made contact with the side walls of the box before landing, we achieve an efficient learning method. Our experimental results demonstrate that the proposed method is possible to learn the tossing motion. Further, we discuss the efficacy in terms of the cycle time and the operating range of pick-and-toss.

As robots are becoming popular labor force, we can expect more robots to work in places more familiar to our daily lives such as restaurants. This paper aims at realizing a manipulation of line-shaped flexible objects to achieve its appetizing looks. As a first step, we try to develop a system for handling and manipulating a line-shaped flexible object: a method for controlling the bias of the object: a method for estimating the center of gravity and contact point to further estimate the distribution of line-shaped flexible objects in the robotic hand. The efficacy of the developed system was validated by several experiences.

Dynamic manipulation of flexible objects such as cloth, which is difficult to modelize, is one of the major challenges in robotics. Humans can move their arms at high speed using their flexible bodies skillfully, and even when the material to be manipulated changes, they can manipulate the material after moving it several times and understanding its dynamics. Therefore, in this study, we focus on the following two points: (1) body control using a variable stiffness mechanism for dynamic manipulation, and (2) response to changes in the material of the manipulated object using parametric bias. By incorporating these two approaches into a deep prediction model, we show through experiments that a musculoskeletal humanoid can dynamically manipulate cloth while detecting changes in the physical properties of the manipulated object.

Robots are expected to manipulate objects with safe and dexterous way. For example, washing dishes is a dexterous operation that involves scrubbing the dishes with a sponge and rinsing them with water. It is necessary to learn it safely without splashing water and without dropping the dishes. In this study, we propose a safe and dexterous manipulation system. The robot acquires the manipulation dynamics of the object by estimating the state of the object and the robot itself, the control input, and the degree of human assistance required (assistance rate) after the human corrects the initial trajectory of the robot hands by interruptive direct teaching. By back-propagating the error between the estimated and the reference value using the acquired dynamics model, the robot can generate a control input that approaches the reference value, for example, so that human assistance is not required and the dish does not move excessively. This allows for adaptive rinsing and scrubbing of dishes with unknown shape and properties. As a result, it is possible to generate safe actions that require less human intervention.

In this study, we focused on odor sampling in the problem of odor source search in a real environment. We aimed to extend the search from one-dimensional to three-degree-of-freedom search in a two-dimensional plane. Furthermore, we engineered the particular behavior of silkworm moths during search.

The upper limb support equipment being developed in this laboratory are intended for hemiplegic patients after stroke and people with reduced muscle strength. We are focusing on EEG to obtain the movement intention. The EEG is useful for rehabilitation of hemiplegic patients after stroke because it is expected to recover the motor function based on the plasticity of the brain. In this study, we aimed to develop a BCI that can acquire the movement intention of a single movement included in a meal movement by using EMG in addition to EEG. We conducted experiments on flexion and extension movements of the shoulder and elbow joints included in the meal movement.

Recently, a deep reinforcement learning approach has been proposed as one of the control methods for legged robots. This paper describes walking motion of a bipedal robot using deep reinforcement learning. In this research, we focused on kinematic synergy in the legs and proposed reward of planar covariation of elevation angles at the thigh, shank and foot during human locomotion. By introducing the reward, the agent was trained, and as a result, the bipedal robot realized forward walking on the heels and off the toes. In addition, the proposed reward is considered to be one of the factors that characterize human walking motion by comparing it with conventional rewards.

Bioactuators, which use muscle tissue or cells as drive sources, have been attracting attention as actuators with unconventional characteristics. It is important for the control of bioactuators to develop a mathematical model, but, to the best of our knowledge, no model has been reported to suppose the application to devices. In this study, Hill's four-element model was applied to tissue-engineered muscles constructed by culturing mouse myoblasts C2C12 in collagen gel. The muscles were isometrically contracted and the maximum contractile force was about 2.9 mN. Then, the parameters of the model were determined from the results of the regression analysis. In addition, the reverse tweezers structure was driven by the muscle. The tip of the reverse tweezers was displaced up to 0.6 mm by muscle contractions. As a next step, we will control the reverse tweezers structure by applying the model.

Passive dynamic locomotion arises only as a result of dynamic interaction with the ground, i.e., without actuators, sensors, or controls. For exploring the principles of human and animal locomotion, passive dynamics is important. One important characteristic is the existence of a stable limit cycle in passive locomotion dynamics. However, to date, no limit cycle has been found in single-legged passive dynamic running. Numerical simulation reveals that the leg dynamics of our 1-PDR model has a stable limit cycle.

Proposed inertial assist method may be useful for passive dynamic biped as well as human walking, where they have a natural self-undulation cycle and basic dynamic stability. An assist can enlarge the forward swing motion of the leg and also a sideways undulation of whole body. However, enlarged motion may lower the original stability. This paper tries to make an adjustment of assist timing and strength to produce a stability by a control based on sensing information. Experimental result shows that assist timing adjustment with laser measurement of swing leg height can make up a stability such as a conversion of stepping cycle time. To produce a large assist torque, a control momentum gyro is introduced on the biped. A cycle time and sideways undulation span is enlarged by gyro system controlled by laser measurement.

In this study, we clarify how the backbone stiffness and the passive aspects of the body influence gait using a QW robot, RW06-Duo, based on a passive dynamic walking mechanism. Some experiments will prove that a knee oscillation with a phase oscillator excites some gaits and gait transition by mutual coordination between the hip and ankle joints, and a torsional backbone. The gait transitions depend not only on the intrinsic frequency of the oscillator but also on the stiffness of the backbone and the passive intralimb coordination. Moreover, we will show the existence of stable gait patterns and gait transition under the same conditions of the phase oscillator frequency and the backbone stiffness.

The supporting force, that legs of the rover is received, is increased when vibration is imparted to the ground. This phenomenon improves the running performance of the legged rovers. In this study, the relationship between vibratory amplitude and increasing the supporting force is investigated. In the experimental results, the more vibratory amplitude is large, the more the supporting force is increased in low sinkage of a leg. However, this result has not been confirmed in high sinkage of a leg. Moreover, the supporting force is correlated with vibratory acceleration which is calculated from vibratory amplitude in low sinkage of a leg. This result means that the supporting force may be estimated from vibratory acceleration. Next, vibratory amplitude and vibratory frequency are compared in order to investigate which is effective for increasing the supporting force. In this results, vibratory amplitude is more effective than vibratory frequency for increasing the supporting force.

This paper proposes integration of a gait control of a hexapod robot and multiple local controls in each leg. A range sensor and linear position sensors equipped on the leg are used in the local controls which are target-pursuit, ground-contact-avoidance, self-collision-avoidance, and obstacle-avoidance. The local controls are integrated based on Riemannian motion policies (RMPs) and are incorporated in follow-the-contact-point (FCP) gait control. The effectiveness of the proposed control architecture is shown in experiments in which a hexapod robot walks over rough terrain.

Various transport facilities and traffic environments for Seamless Mobility are being developed worldwide. Most of those project require large-scale infrastructure development, and it is difficult to carry out it in rural towns and unknown environment. Therefore, another solution which doesn’t depend infrastructure development is ultimately required. As a solution, we develop a six-legged robot named ”SOL” for seamlessly navigating various terrains. SOL moves with discrete event-driven control. In this paper, contact assistance method which put SOL’s leg tip in close contact with the ground is proposed.

Wall-climbing quadruped robot is developed based on the Universal Vacuum Gripper (UVG). UVG is one of suction cup adhering to uneven surfaces with its deformable lip. The robot developed in the previous research has a simple structure consisting of two legs equipped with UVG and two degrees of freedom of rotation, and was able to climb uneven walls and move along the ceiling, but could not perform unattended inspections due to the limited degree of freedom. Therefore, we attached UVGs to the four legs of a quadruped robot for automatic inspections of social infrastructure. We reports the current developmental status of the wall-climbing robot.

Rescue robots used on the ground include wheel-type, crawler-type, and leg-type robots. Each robot has its own strengths and weaknesses, so they are used in combination. However, it is difficult to separate various environments in rescue activities. In other words, since the environment that is the strength of the robot does not continue, activities in the environment that is the weakness are inevitable. Therefore, it is necessary to take measures for the weakness environment. In this research, we focused on a bipod robot that is not good at muddy environments. Then, we developed a bipod robot that is supposed to be operated in a muddy environment.

To date, many robots have been researched and developed for the purpose of investigation and support at disaster sites, and quadruped robots are expected to play an important role because of their superiority in traversing rough terrain and stability. At present, there are many quadruped robots driven by electromagnetic actuators for investigation and inspection, but there are few quadruped robots suitable for heavy work and carrying heavy objects. The purpose of this study is to develop a hydraulic quadruped robot that can perform heavy work even in a severe environment by utilizing the features of hydraulic actuators. As a result, the potential of the field of hydraulic quadruped robots will be opened up.

n recent years, various researches on humanoid robots have been conducted, and their practicality has been increasing. However, bipedal walking, which is the main means of locomotion for humanoid robots, has several problems. The first is the slow movement speed. The second is that it is not energy efficient. In order to solve these problems, we devised a skating method for robots. However, at present, the speed is slow and constant. By increasing the opening angle of the legs, we can change the propulsive force. By using this method, we aim to increase the speed of the robot and change it arbitrarily.

Until now, many types of bipedal leg-wheeled mobile robots have been proposed. However, a compound locomotion, which combines wheeled locomotion and walking simultaneously, have not yet been studied enough. In this paper, the compound locomotion method is proposed in 2D sagittal plane using an optimization technique based on quadratic programming and the effectiveness is demonstrated through numerical simulation results.

Legged walking animals can stabilize their posture without falling down using only a few legs. In this study, we aim to reproduce and understand the staggering motion, a strategy to recover posture by moving the legs as quickly as possible in response to a change in posture. In order to reproduce the staggering motion seen in animals and humans, this paper reports on a robot that performs staggering motion by using a brainless approach in which the entire robot body is controlled by pure mechanisms such as valves and switches, without using any kind of computation such as animal brains.

We have been continuously developing a hexapod walking robot based on the mobile form of a harvestman because its mobile form consisting of long legs and a small body is advantageous for locomotion on complex uneven terrain. In this report, we showed the concept of second prototype of a harvestman-like hexapod walking robot ASURA-II first. A new design principle called "double-coupled drive" has been introduced to the drive system of the leg mechanism to achieve a long leg structure and a wide range of joint motions. Next, the details of the design of the leg mechanism using double-coupled drive, chemical fiber wire, and the principle of dynamic pulley were described. Finally, the leg mechanism analysis conducted to obtain design guidelines were reported.

SLAM based on map generation of the surrounding environment and self-position estimation is effective for multi-legged robots to demonstrate its performance. Most of the previous SLAM researches focused on odometry using wheels, and there are few examples of verification using a multi-legged robot. In addition, due to the limited payload of a multi-legged robot, it is necessary to perform processing with limited sensors and onboard resources. In this study, a SLAM method using a small quadruped robot, a depth camera that can be mounted on the robot, and its built-in IMU sensor is proposed. A method to estimate the position and posture of the robot by mapping feature points obtained from color images between image frames and distance images is implemented, and evaluated.

In recent years, the use of 3D printed plastic parts has been attracting attention. However, due to their inferior dimension accuracy, stiffness and strength, there are few examples of using them as structural parts for robots that are subjected to loads. In this study, we developed a quadruped robot, TITAN-E1, in order to verify the applicability of plastic parts made by fused deposition modeling 3D printers to robot structural parts. As a result, it was found that the 3D printed plastic parts had the sufficient dimensional accuracy and the stiffness and strength required for stable static walking. Therefore, the applicability of the 3D printed plastic parts to the structural parts of the quadruped robot was confirmed.

To realize a strong and simple structured legged robot that continues walking after turning over, we have developed an all-side multiple-legged mechanism driven by a single motor. The mechanism comprises a driving worm shaft and multiple worm gears, which drives legs at all side directions by a pseudo leg linkage mechanism. Setting initial phases of all the legs to satisfy the solution of graph-coloring problems on a cylindrically connected graph, it was shown that the mechanism realizes a tripod-like gait at every direction. In previous reports, we used floating crank linkage and pantograph mechanism as pseudo leg movement. However, these pseudo legs were insufficient in realizing deep radial stroke. It was, therefore, difficult to perform stepping over action. This report shows a new pseudo leg linkage that gains roughly twice as high as the previous designs. Also, we show a driver design that integrates a coaxial planetary gear reduction mechanism.

We have been proposing a passively driven gripping foot that can be attached to any mobile legged robot, which enables the robot to move in rough terrains with uneven surfaces. In the previous report, we have shown a prototype foot design that grips bumps or rocks on a floor from side by nails pushed by the robot’s weight. In this report, we show the relationship between the characteristics of walkable terrain and the design parameter of the foot. First, we prepared a small experimental stony slope. Then, we measured the average roughness of the surface on the slope by using 3D measurement from the reconstruction of multiple images. The performance of the robot to walk by the foot was tested on various inclination of the slope. As a result, it was found that the diameter of the proposed legs should be set to about 4 times as large as the average roughness of the surface to climb on the slope up to 7 degrees inclination.

We are developing a four-legged robot with two active joins in both forelegs and two active and one driven joint in both hind legs. In this report, we present a new soft sensor for detecting the ground contact. Previously we used a film-shaped pressure sensor, and a method a mechanical switch with a small ball, which were installed in the leg’s toes to detect the toe’s contact timing with the ground. However, they did not work sufficiently. Therefore, we designed and fabricated a new sensor, which is balloon-shaped silicone rubber and has a pressure sensor to detect the change of air pressure inside the balloon caused by the toe’s ground contact. As a result of the preliminary experiment, we confirmed that the new soft sensor can detect the ground contact.

If the legs of a quadruped robot have elasticity, it will be able to walk like an animal. In this paper, we design a control system of a quadruped robot with elastic legs to realize trot walking. We use a van der Pol equation as a walking pattern oscillator to generate a trot gait pattern of the robot. Simulations and experiments are performed to find the conditions for a stable trot walking and a trot gait pattern that maximizes the walking speed.

In this study, we developed a very simple walking assist suit based on principles of passive walking and human walking. We evaluated the walking assist suit by questionnaires and experiment of double pendulum. In this paper, in order to clarify effects of the walking assist suit, we conducted walking experiment by attaching the walking assist suit to 2D passive walker.

Stretchable optical waveguides provide viable sensing solutions for soft robots. So far, basic single layer or core-cladding waveguides have been developed using various manufacturing methods. Extending their structure to increase sensing functionality has been mostly unexplored. This study investigates a waveguide design featuring a semi-divided core cross section guiding the light from one light source to two light outputs. With optical ray tracing simulations using COMSOL, we verify the design’s two main working principles that enable two-degree-of-freedom sensing in various strain modes.

The integration of truncated conical actuators, whose radius can be actively controlled, realizes a surface robot that can bend both sides. We call this concept an S-isothermic surface robot and have realized a new pneumatically driven robot. We applied a simple pressure pattern to the robot and measured the robot’s deformation using a motion capture system. From the deformation results, the curvature and pseudo-isometric error will be discussed.

This manuscript focuses on deformation modeling of viscoelastic materials. Analysis of dynamic behavior of soft robots requires dynamic modeling of soft robot materials. We formulate dynamic deformation of viscoelastic materials via power laws. Model parameters in power laws are identified by numerically minimizing the error between measured and calculated stress–strain relationships.

Soft robots have advantages in terms of safety, softness, and compliance compared to traditional robotic systems. However, soft actuators, often employed in soft robots, require a corresponding number of bulky pressure supplies/valves to drive. Here, we consider a valve that can control the flow without mechanical moving parts for simplifying the driving system of soft actuators. We developed a system comprising a pressure source, a switching valve, and two actuators to demonstrate the feasibility of introducing a fluid valve into soft robotics. As the valve, which makes use of the Coanda effect, can switch the flow between two outlets when the pressure difference between the outlets is 2 kPa, we employed a soft artificial muscle actuator connected to each outlet. The system can control the shrink of each actuator by switching the control flow. The experimental results proved that the system could actuate each actuator at will.

The soft gripper is possible to fit in the shape of grasped objects and realize safe handling due to its softness. On the other hand, modeling and control of the grasping force by the soft gripper has not been focused yet on so far even though it is also important for not injuring them. In this study, a sensorless tip force estimation of the cable-driven soft gripper is proposed because of the difficulty of implementing sensors to the soft body generally. The model to estimate the tip force is derived from the dynamics of the soft gripper based on the lumped parameterized system. Its effectiveness and limitation are demonstrated through numerical simulation results.

In this study, we develop a tensegrity manipulator driven by 40 pneumatic cylinders without mechanical springs. In general, tensegrity robots use mechanical springs to achive a stable and curved tensegrity structure even when a component is actively extended and re tracted by an actuator. The stiffness of the mechanical spring should be high to increase the stiffness of the entire structure and imporve the control response, but low to deform the structure. This means that the implementation of mechanical springs causes serious trade-offs in its design and contorl. In this study, we use pneumatic cylinders not only for active deformation but also for passive deformation to realize a tensegrity robot without mechanical springs. In this paper, we introduce the robot’s design, control system and the result of posture change experiment.

The purpose of this research is to develop an elephant trunk shaped power soft robotic arm with both high power and flexibility. Our proposed arm consists of three elements: hydraulic artificial muscles to drive the arm, a compressive stress support component reinforced with fabric, and a gripper reinforced with knit. In this paper, We fabricated prototypes of the compressive stress support component and the gripper. Then their performance was confirmed by experiments. The experiment confirmed that the stiffness of the compressive stress support component can be changed by varying the applied pneumatic pressure. And the experiment also confirmed that the gripper can grasp objects of up to 20 kg.The purpose of this research is to develop an elephant trunk shaped power soft robotic arm with both high power and flexibility. Our proposed arm consists of three elements: hydraulic artificial muscles to drive the arm, a compressive stress support component reinforced with fabric, and a gripper reinforced with knit. In this paper, We fabricated prototypes of the compressive stress support component and the gripper. Then their performance was confirmed by experiments. The experiment confirmed that the stiffness of the compressive stress support component can be changed by varying the applied pneumatic pressure. And the experiment also confirmed that the gripper can grasp objects of up to 20 kg.

This paper describes a soft robot that aims to move forward in an unknown space as it recognizes and adapts to the spatial shape of the environment. It was previously reported that soft tensegrity and Recurrent Neural Network (RNN) can be used to realize tensegrity structure shape recognition. In this study, a tensegrity robot was designed to actively generate propulsive force as it presses its body against a wall in its surrounding environment. This robot design includes a novel artificial muscle arrangement called ”4/3 muscle winding,” which induces large deformation in the tensegrity structure. This tensegrity robot could enter an unknown space, such as a cave, and recognize the spatial shape of the surrounding environment by recognizing the tensegrity structure shape.

Soft robotic is a rapidly developing field exploiting biomimetic design principles, novel actuation concepts, and manufacturing techniques. The present study focuses on the development of a spiral pneumatic rubber actuator which can transport objects placed on the surface of actuator. The actuator consists of multiple air chambers and provides sequentially traveling waves when the pressure is applied to successive chambers. The fabrication process and mechanism of the spiral pneumatic rubber actuator are discussed in this paper. Throughout the experiment, it is found that the sequential deformation of successive chambers could generate traveling waves on the surface, thus achieving transportation of the objects.

Ionic Polymer Metal Composite (IPMC) is planar and it is necessary to apply AC voltage to both sides of the IPMC in order to drive it repeatedly. Since the generated force is small, the interference caused by the wiring, critically limits the motion. In this study, I consider the following two issues separately in order to drive an independent IPMC robot integrated with a battery. First, to make it wireless, a magnesium battery is installed. Second, a spiral IPMC is used to drive the robot only by the battery characteristics. As a result, I confirmed that the spiral IPMC can be driven only by the characteristics of the magnesium battery. In the future, I will design a structure that enables even greater motion, and fabricate a self-contained IPMC robot that incorporates the above elements.

Independent walking has always been an important issue in the lives of visually impaired persons, and the limited visual information of surrounding environment is the key to making it difficult for visually impaired persons to walk independently. In order to solve the defect of visually impaired persons acquisition of surrounding environment information, we choose to use sound information to make up for the problem that visually impaired persons cannot rely on visual information to judge the surrounding environment when walking.Therefore, we prototyped a maneuvering system using force sensors based on sounds from a SLAM-equipped wheeled mobile robot. This report describes the system configuration and basic experiments.

Ceiling is considered to be suitable location for robots to transport cargoes because Automated Guided Vehicle(AGV) and people do not interfere with them. In our previous study, we have developed a ceiling mobile robot called HanGrawler 2. HanGrawler 2 can travel at a high speed of 1.0 m/s to compete with ground vehicles as its champion data. However, HanGrawler 2 sometimes fell at the beginning of the traveling. So, the purpose of this study is to improve the traveling stability of HanGrawler 2. There are two technical issues to be solved: 1) HanGrawler 2 cannot start traveling until hanging mechanism is set to specific position, and 2) HanGrawler 2 tends to fail to hang to the ceiling during acceleration. Automatic adjustment of an initial position and new control method of hanging considering in acceleration is added to solve these issues.

In this research, we develop a small wheeled inverted pendulum robot, μ-Pentar, equipped with a Raspberry Pi 4B. Due to the shortage of semiconductors in this year, we were unable to obtain an FPGA which was to control the motor driver and perform the signal processing for the encoder. Therefore, we investigated a method using the pigpio library instead of the FPGA. In this paper, the results of measure the response time of encoder counter function of pigpio library and the latency of serial communication line for IMU that measures the posture of the robots.

We developed a soft pneumatic actuator that can bend and deform with a large curvature, mimicking the deformation of a pill bug by the folded structure. This paper describes the design using a structure that folds and deforms by using deformable materials and optimizing the folding part to minimize the bending stress generated during folding. Driving the actuator with vacuum pressure helps obtain fail-safe characteristics and constrain the crease’s bending angle. It has been possible to drive the actuator with a vacuum pressure acting uniformly in the flow channel by increasing the stiffness of the parts not involved in the deformation.

This paper proposes a method for a continuum robot to achieve a desired complex shape (e.g. Clothoid curve, Lemniscate ) that cannot be modeled using a constant curvature model. This can be realized by freely determining the lengths between discs on the continuum robot. We developed a simulation that reproduces the predefined target shape by optimizing the deformed robot shape to be close to the target shape. The proposed model can be utilized for designing a manipulator, a body part of a biomimetic robot, or a robot hand.

Flexible manipulators have advantages such as collision safety, the ability to complete subtasks, and the realization of rational motion patterns. However, their mechanisms are too complicated to be controlled. The use of an underactuated drive system is a common solution to overcome such a problem. However, an underactuated manipulator hardly realizes rational motion patterns, high-acceleration motion, or gravity compensation. Although these features are easily realized in a vertebrate. In this study, we focused on the motion of an ostrich that skillfully manipulates its flexible neck, and developed a manipulator based on anatomical knowledge. The developed manipulator was very flexible, and we can raise its head against gravity with a neck motion pattern similar to that of an ostrich. The results of this research provide a suggestion to potentially explore muscle synergy during ostrich neck motion.