In this study, we improve a control method for a multi-robot system to control network bi-connectivity. It is desirable for the network of a cooperative multi-robot system to be connected since the robots in the system exchange information via the network. To withstand a failure of robot, bi-connectedness of the network is important. The bi-connectedness should be resumed when it is violated. Thus, some control methods to preserve and resume the bi-connectivity are proposed. However, there exists some configurations with which the bi-connectivity is unrepairable by such control methods. To overcome the difficulty, we propose a method in which robustness of the connectivity is temporally abandoned. The network structure becomes robust against robotic failures by our method in the long-term perspective, although it may be fragile in the short-term. Some results of computer simulations are shown to verify the proposed method.

In this study, we develop a multi-robot experimental system and experimentally verify a network connectivity preservation method. We introduced a distributed algorithm for estimating an eigenvalue of graph Laplacian referring to previous studies and conducted control experiments on network connectivity preservation. In this paper, we report the details of the eigenvalue estimated in the distributed manner and the results of experiments on the network connectivity preservation control.

In order to design a controller for cooperative transportation by swarm robots, there is an approach to optimize a Neural Network (NN) as a controller using reinforcement learning. In most of the previous studies, the NN used the full coupling layer, which cannot cope with changes in the number of robots and states observed by the robots. In this study, we adopt a NN with the Self-Attention mechanism proposed in machine translation as the controller of the robot. Since this mechanism can respond to changes in the number of inputs, it has the potential to cope with environments where the amount of state observed by the robot is different from that during training. In this paper, we train the cooperative transport system in an environment with three robots and three obstacles, and verify whether the system can respond to changes in the number of robots and obstacles using the controller.

Self-assembly is an important factor for swarms to provide robustness to the system. For swarm robot systems, robots have been considered to realize self-assembly like ants and to adapt to their environment. These robots, however, have been modeled based on living organisms, and there are few studies that have further developed self-assembly in artificial swarm robot systems. This paper proposes a connected swarm system for artificial self-assembly. Firstly, we develop the system in which the robot can switch its driving mode from active to passive, and vice versa.

Ants do not act complicatedly as individuals, but when viewed as a group, they work together to protect themselves from foreign enemies and take cooperative actions such as collecting food. They have classes; each ant that is assigned a role cooperates with others and creates a society to prosper the species. Such a system that each element acts autonomously without receiving any information from a central function that integrates the whole but it functions as a whole due to the interaction with others is called an “autonomous decentralized system”. In order for multiple robots to perform cooperative movements in a changing environment, it seems effective to make the robots acquire their own abilities and mutual relationships between them in a learning manner, regarding the whole as an autonomous decentralized system In this study, we evaluate the dynamic characteristics of mobile robots that were designed and prototyped to conduct research.

Today, many robots such as industrial robot arms and automatic agricultural machines work operation that is simple or put a burden on humans in various places instead of ourselves. However, in the present circumstances, they are good at repeating motions instructed by computer programs, whereas they cannot handle unexpected circumstances easily; so the level of a robot which can actively and creatively act continues to be an important research subject. Besides, it is expected to make "multi robots" perform a certain task cooperatively under a changing situation. Especially from the latter point of view, it seems to be effective to make the robots acquire their own abilities and the mutual relationships between them in a learning manner, regarding the whole as an autonomous decentralized system. In this research, we designed and prototyped some moving robots (four-wheeled vehicles) and evaluate their basic characteristic in order to conduct our actual machine-based research.

Shepherding problem is to guide a large number of moving objects (sheep flock) with a small number of controllers (shepherd, sheepdog). Exploring the shepherd guidance model is useful for controlling multiple robot and guiding real swarm of organisms. In this paper, to demonstrate sheepdog system in the real world, we proposed a control system for a sheep robot and a sheepdog robot based on a two-wheeled vehicle model. Through simulation validation, we have verified the validity of the proposed system.

Swarm robots have attracted great attention due to characteristics such as high robustness, adaptability, etc. Among them, the reconfigurable modular robot is a type of swarm robot formed by combining repeating robotic modules to execute functions that neither scattered swarm robots nor individual robots can. However, these modules might lose some of the flexibility of swarm robots because the connections between these modules are strong. Therefore, in this research, we propose a swarm robot concept called “ Weak Constraint Swarm ”which connects robotic modules by “weak constraints” to obtain functions that cannot be achieved by individual robots, dispersed robot swarm, or modular robots. Then, we propose a swarm robot system called“ ICT-Sheet ”based on the“ Weak Constraint Swarm ”concept for navigation problems in rugged mountainous terrain. And the value of the“Weak Constraint Swarm” approach is further validated by demonstrating the unique capability of“ICT-Sheet”for this particular navigation problem in an experiment.

This paper presents a novel tube-type pneumatic helical actuator generating complex 3-dimensional torsional motions. Inspired by the molecular structure of DNA, this actuator is fabricated by combining two helical contraction actuators, introduced in our previous research, in parallel and covered by a sleeve. By controlling the air pressure and top rotation angle of each helical actuator, this double helical actuator can realize bidirectional screwing, curving, and mixture of screwing and curving in a 3-dimensional space. Qualitative working principles of these patterns are analyzed, and two mathematical models are introduced to explain the motion patterns of this actuator, and validated through experiments. The effectiveness as an application to a flexible manipulator for inspection in narrow spaces is investigated and its validity is experimentally verified.

This paper presents air cylinder type artificial muscle and measurement of its output characteristics. First, the characteristics of conventional air cylinders are listed and some issues are discussed. To solve these problems, we propose a small diameter flexible air cylinder that can be placed and operated in a bent position. However, a small diameter flexible air cylinder is difficult to control because of its small output and large variation in output. The output characteristics of a small diameter flexible air cylinder and an artificial muscle made of a bundle of small diameter flexible air cylinders are measured. The results are compared with those of conventional air cylinders and discussed.

Recently, soft growing robots have attracted great interest. They move with extension from their tip and making bending shape, and they can move in clutter environments. Many soft growing robots and our previous robot were composed of an inflatable structure. The stiffness models of a straight inflatable beam have already developed, but they are not enough for soft growing robots with an inflatable tube because they have some bending shapes. In this paper, we developed deformation models of a bending inflatable tube and compared them to the actual deformation of an inflatable tube. We confirmed that an inflatable tube bent around the pitch axis can be regard as a torsion spring and our deformation model agrees with experimental results. These models enable us to calculate the bending angle and the height that the robot can reach with pitch-axis bending.

In this study, the passive jumping mechanism based on the jumping principle of click beetles is proposed. The proposed mechanism consists of two links, a peg made from viscoelastic material, and springs. When an external force is applied to extend the joint, the joint is temporarily fixed in an extend state. Then, If the mechanism is placed on the ground and wait for a few seconds in this state, the joints will be released and the mechanism will jump. In this paper, we fabricated a prototype of a passive jumping mechanism and it was confirmed that the proposed mechanism is possible to jump in passively.

Soft robots have been developed in recent years. In order for a soft robot to perform complex motions, it is needed to operate multiple actuators using multiple wires. Multiple wires increase the weight of the soft robot and limits the degree of freedom. We propose a system that uses a single channel to operate a specific actuator from a group of them by combining conductive droplet spacing and electrode spacing. In this system, two electrodes facing each other across the channel are placed at different intervals to connect the actuator to the power supply. While flowing insulating fluid in the channel, multiple conductive droplets are injected at various intervals. Regardless of the order in which the actuators are placed, only those actuators with electrodes that have the same intervals as conductive droplets will operate. This method can simplify the driving system and reduce the weight of soft robots.

“Nimbus Sheet”, the envrionment manipulation mechanism with free adaptability has been developed with soft cilia vibration. To realize the function of transporting a heavy load object, a top plate as supporting mechanism was adapted. This top supporting plate can reduce the deformation of cilia that decreases the transporting capability of Nimbus Sheet. In addition, we verified that the weight capacity can be improved by controlling the contact state of the cilia with transporting objects.

In this study, “Nimbus Sheet”, a flexible non-grasping manipulator with soft cilia that can be adapted to various human environments has been developed. We fabricated cilia by using soft materials and vibrated them with voice-coil actuators. With a super-slow camera to capture images of the object being transported, we were able to clarify the transport principle of this device. By changing the frequency of the input signal to the voice coil actuator and measuring the propulsion, we were able to confirm the transport performance of the device.

In this paper, we propose a toroidal origami monotrack for circulating and bending of closed skin along origami bellows folding structure. Conventional whole skin drive methods achieve circulating motion, however, the mechanism to prevent occurring unexpected wrinkles with smooth forward motion and bending motion is unclear. Here, we introduced a mechanism to drive the skin along the bellows structure which friction on the edges is reduced by rollers. The proposed system may apply to robotic tasks where complete airtight is required, such as sterile rooms, sewage treatment rooms, and other extreme environments. The prototype track test showed the potential to realize a smooth closed skin drive locomotion.

This study proposes the analysis and validation of the novel mechanism that can switch the airflow path by a single airflow control for the pneumatic soft robotic hand with the variable friction mechanism. The variable friction mechanism functions by the pneumatic driven, using the Venturi effect. By the flow-channel-switching (FCS) mechanism, the proposed robotic hand achieves the grasping motion and friction-reduction function through the 1 DOF airflow control. In this paper, the condition of the design parameters to realize the desired behavior of the FCS mechanism is presented through the mechanical analysis. The analysis is validated through the actual experiments.

Textile is soft, adaptable to shape, elastic, and lighter than metal. Taking advantage of these characteristics, textile is used as a material for soft robots. The authors thought that by using a cloth made of high-strength fibers, it would be possible to create a robot that takes advantage of its strength and durability in addition to its flexibility. As the first step to use textile as a mechanical component of a robot, this paper proposed a new evaluation method and a method to implement a testing machine. This paper also conducted strength and durability tests of high-strength textile using the testing machine.

This paper presents joint with elastic cord and measurement of joint stiffness. The flexibility of robot joints is one of the factors required for human-robot coexistence. In order to make the joints flexible, we focused on flexible strings and listed the advantages of using strings as joint structures. Thus, we propose a joint with elastic cord. The structure of the joint with elastic cord and the structure of the net composed of the elastic cord used for the joint are described. The fabricated joints with elastic cord were measured. From the experimental results, the relationship between the structure of the net and the stiffness against the direction and angle of joint flexion was discussed.

An ionic polymer metal composite (IPMC) is a polymer actuator that bends when a voltage is applied. In this paper, we proposed a method for producing a three-dimensional IPMC by three-dimensionally molding an ion exchange resin using a water-soluble support material and plating it with gold. The proposed method has made it possible to manufacture IPMC robots with a complicated three-dimensional shape more easily than before. We showed that IPMC can be manufactured using the water-soluble resin PolyDissolve by the proposed method. We showed that the performance of the completed IPMC changes depending on the solvent component of the Nafion dispersion solution used for molding the ion exchange resin. We also showed that an IPMC robot with a complicated three-dimensional shape can be manufactured using the proposed method, and confirmed its operation.

This study aims at fabrication of a soft active tactile sensor with three chambers including strain gauges, which can change its sensitivity by changing air pressure inside the chambers. In this report, we fabricated two types of tactile sensor. In the one model, the three chambers were connected, and in the other, they are independent from each other. We used a stereolithographic 3D printer to fabricate the sensor’s base structure with three chambers inside the body. We investigated how the difference between the two structures would affect the strain gauge outputs. In our experiments, we varied the gauge pressure and slide an indenter on the surface of the tactile sensors. From the experimental results, we confirmed that connecting the chambers increases the sensitivity but generates extra output, and that the sensitivity changes depending on the value of the air pressure inside the chambers.

In this paper, we describe a robot for inspecting concrete walls that has suction feet and can overcome steps. The inspection of concrete walls is necessary for the maintenance and management of concrete structures. However, it requires a lot of preparation and is dangerous to be done by humans. Therefore, the use of robots is required to solve these problems. A robot that can move through walls should be lightweight. Therefore, we have developed a wall climbing robot capable of stepping oversteps with a reduced degree of freedom. This robot has one degree of freedom for each of its two legs and can turn and move up and down. it moves by rotating its two legs alternately.

A mobile robot that moves in narrow environments, such as inside a pipe, usually uses a crawling mechanism. Most crawling mechanisms use an inflation part that anchors the body onto a wall and a contraction part that drives the rest of the body forward. However, the infraction part usually uses a rubber air expansion bag or a complex wall pushing mechanism, which are not suitable for extreme environments. For example, a rubber bag may easily be damaged by a hot temperature. Also, a complex wall pushing mechanism may not be used in narrow pipes. To make durable and simple crawling robots, we have proposed a mechanism that only uses twisting motion. The motion drives a body forward while thin plates inflate to anchor the body. In our previous report, we introduced a principle of this movement. We also showed a prototype robot that has a limited sheet inflation diameter. This report proposes a mechanism that realizes limitless inflation diameter by winding plates multiple times around the body. We also show a new prototype that includes batteries and controllers.

In this study, in order to clarify the mechanical relationship of forces and moments acting on an U-shaped suspension type robot climbing a cylindrical column, we examine methods to measure the force received from the cylinder acting on the robot’s wheels using a 6-axis force sensor. This robot supports its body with ropes and its own weight, like an U-shaped suspension used by column workers to prevent them from falling, and moves through the column in a spiral path with the driving force of its wheels. In order for this robot to achieve stable lifting and lowering motions, it is crucial to know the magnitude of the pushing force of the wheels and rope against the cylinder. In this paper, we propose a method to measure the force acting at the contact point of the wheel and cylinder with 6-axis force sensor, and show the validates of the method experimentally.

Wireless power transfer via magnetic resonant coupling can be used to supply power to a mobile robot within a few meters of a transmitter coil. However, when the robot moves or its power consumption fluctuates, its input impedance varies and causes power reflection. Therefore, we propose the use of multiple identical driver coils on the transmitter side in order to match the input impedance. The input impedance is matched and power reflection is eliminated by switching driver coils, i.e., regulating the coupling coefficient between the driver and the transmitter. In addition, the winding number of the receiver coil can be changed to relax the design condition of the drive coil. The experimental results showed the input impedance was well-matched under a wide range of distances and load resistances. Therefore, the proposed system was expected to solve the power reflection issues in mobile robots automatically.

Snake-like robot is a hyper-redundant mobile robot with an articulated body composed of serially-connected modules driven by actuators. Recently several hyper-redundant mobile robots have been developed in hope of the versatility brought from its simple form. Snake-like robots have been developed aiming at its applications to operations dangerous to man: the in-pipe inspection at chemical or nuclear energy plants and the rescue of victims under collapsed buildings by making use of its flexible and slender shape, and the mine detection by distributing its own weight on the ground and so on. In this paper, the novel mechanical design is proposed to enhance the mobility for various environments, and a prototype model is composed of serially-connected modules with suspensions. The performance of prototype model is described

In this study, we aim to improve the mobility of the capsule type robot by generating horizontal and vertical motion patterns of inner mass. The motion patterns of the inner mass are generated using Bezier curves, and the control points of the Bezier curves are determined by numerical optimization. The model of the capsule type robot is constructed, the motion patterns of the inner mass are generated, and the results are verified by simulation. As a result of the simulation, it is confirmed that the of movement of the capsule type robot is improved by using the proposed method of the inner mass.

A wide variety of species living in the world have unique physical functions and tissue structures that have evolved to survive in the harsh natural world. These features that humans cannot possess have been reproduced by technology to improve performance and expand the functions of machines. Mobile robots that imitate the characteristics of living organisms use a variety of mechanisms, such as wheeled, legged, snake-type, and peristaltic types. In terms of the range of motion and frontal projected area, the peristaltic type is superior for moving in a conduit. In addition, the peristaltic type is expected to be useful in narrow spaces in view of its ability to bend. In this study, we have developed a peristaltic mobile robot with a kinematic cable drive mechanism using three compression springs.

The omnidirectional motion can move any direction without changing the direction of its own. The authors have developed Wave-wheel mechanism and Screw-type differential rotating mechanism. Also, the stepping performance were improved by extending the ground point from point to line. In addition, we have evaluated the motion characteristics of one ciliated wheel mechanism. In this paper, we examined the composition for vehiclization. In the vehicle configuration, the minimum necessary two units are arranged in parallel. We conducted experiments using actual mechanisms and confirmed that the inventive principle is also effective for vehicle construction, omnidirectional motion.

Snake-like robots are highly adaptable to environments where it is difficult to move in with feet or crawlers. Most snake-like robots perform a “Lateral Undulation” in which they move twisting their bodies from side to side. This method enables them to move over bumps and narrow gaps , but have disadvantage of that is difficult to move on sandy soil or snow. In this paper, we focus on the “sidewinding propulsion method”, which can move in frictionless environments, and propose a moving mechanism for it. In this study, the sidewinding propulsive motion is reproduced by lifting the trunk in conjunction with the bending of the fuselage using a screw shaft, while reduced degrees of freedom.

In agriculture work, weeding work in paddy field has been physical burden and taken a lot of working time. As a solution of this problem, stopping the growth of weed by stirring the soil is affective. An example is "Duck farming" that ducks stir the mad by swimming in paddy fields. In this paper, we propose new robot that imitate snakes to stir the soil. A body segment of the snake-like robot is composed of three bending modules , continuous joint mechanism, layered gear and two wire driving mechanisms, we approximate the serpenoid curve by curves without using straight line. we have developed a prototype of a serpentine robot with a bending mechanism that generates three different curvatures inside one module by wire driving to reduce the degree of freedom. This bending mechanism makes it possible to approximate the serpenoid curve, which is the motion curve of the snake, in terms of curvature. Since the inside of the module unit needs to be bent with a certain curvature in order to derive the posture, a double joint mechanism is installed inside.An operation experiment was conducted using the developed prototype. also, We evaluated the approximation accuracy of the curvature approximation based on the airframe characteristics of the developed module, evaluated the unit ratio inside the module, and confirmed the effectiveness of the intermediate unit.

It is widely known that robots using McKibben pneumatic actuator can achieve various dynamic motions by simple control, such as adjusting the timing of pressure application. However, the design of the applied pressure is often determined by trial and error, and there is no systematic design method. Our previous study proposed a control method for musculoskeletal robots using tension feedback cooperative control. We also proposed a trajectory-following control method for the robots. This paper proposes a new control method that autonomously generates input pressures to realize a target trajectory by combining these two control methods. Simulations and experiments verified the function of the proposed control method. From the simulation and experiment results, we confirmed that the pressure inputs changed autonomously according to the period and amplitude of the target trajectory by appropriately selecting the weight coefficient. In addition to realizing the target trajectory, it was also confirmed that the generation of excessive tension was suppressed.

This study proposes a flexible power-assist suit using a hydrostatic skeleton driving mechanism. For the purpose of a flexible, safe and slip-on system, the power-assist suit consists of a commercially available working cloth and a flexible actuator system. With this system, a dedicated handy controller is newly manufactured, and the power units such as compressor and battery are installed within a backpack to realize a portable system. The suit supports the motions of upper arm lifting, the elbow expansion and contraction. This paper describes a prototype suit system and its experiment. The total weight of the system realizes extremely light weight, approximately 3 kg. Moreover, the suit has achieved all the proposed motion.

McKibben Pneumatic Actuator(MPA), one of the soft actuators, has various useful features as an actuator for robots. Therefore, many dynamic motions by a robot using MPA as an actuator have been realized. The tension of MPA is affected by the mechanical and geometric shapes of the robot’s body because the tension of MPA depends on its length. However, few robots have been designed with this effect in mind. In this study, we examined the effect of the elliptical patella on the locomotion of a legged robot that performs a jumping motion. Genetic Algorithm(GA) was used to optimize the shape suitable for the jumping motion, and it was confirmed that the characteristics of the optimized shape were consistent with the hypothesis.

In this paper, we propose a configuration of a muscle actuator that can reproduce human muscle characteristics. We also propose a control method for the output of the actuator. We confirm that this control method can reproduce the muscle characteristics of two subjects that are from person to person in experiments.

This paper shows directional steering method of In-pipe robot in branch pipes, especially vertical ones. To realize this method, we need to meet 2 conditions. One is geometrical condition of curving shape to insert head of robot into brunched destination, the other is mechanical condition of flexural rigidity not to buckle in branch pipe. In order to meet these conditions, we propose new directional steering unit for geometrical condition and rigidity improving structure for mechanical condition. Directional steering unit can generate bending motion with curvature difference, so it can adapt shape of brunch pipe. Rigidity improving structure can improve rigidity of unit without impeding motion of units because of cylinder of the structure. By these methods, prototype robot can propel vertical branch pipe.

A musculoskeletal model with McKibben pneumatic actuators generates locomotion by an antagonistic drive of the actuators as real animals. However, the coordination pattern of these antagonistic actuators have been determined by trial and error. Our previous study proposed a control method for musculoskeletal models using “tension feedback cooperative control law”, in which the antagonistic actuators cooperate autonomously. This study applied this control method to a legged model with antagonistic muscle groups. We verified the autonomous coordination of intra- and inter-muscle groups and the generation of periodic motions by simulation. The simulation results confirmed that the control method generated autonomous coordination of muscle groups and periodic motion of the leg. It is also confirmed that different cooperative modes are generated depending on the of feedback gain value, and the number of such cooperative modes increases than in previous studies.

Straight-fiber-type pneumatic rubber artificial muscle (SF-PAM) with passive flexibility is a safe and friendly actuator for humans because it can construct back-drivable mechanisms. In conventional antagonist muscle joint structures using circular pulleys, there is a trade-off between range of motion and output torque due to the size of the circular pulley diameter. In addition, the output torque of SF-PAM is insufficient when the arm rotates to some extent due to its output characteristic that the contraction force decreases with contraction. Therefore, by replacing the pulley with a non-circular pulley, we attempted to improve the output characteristics compared to the conventional joint. First, we demonstrated the effectiveness of the non-circular pulley by a torque comparison experiment using a single SF-PAM. Then we modified the existing model equations to accommodate the non-circular pulley and conducted a positional control experiment on the antagonist muscle joint structure using the new control system.

We focus on the changes in functions such as variable stiffness and actuation obtained depending on the input to fluidic core elements enclosed in the structure. In this study, a bellows is enclosed in a linear structure as a fluidic core element. This enables extension and contraction in addition to variable stiffness, bending. As a result of measuring the contraction force and the torque for holding the joint angle, a torque of 0.016 Nm was obtained at 35 kPa. It was confirmed that the measured and the theoretical value of holding torque agreed. If the joint angle was horizontal, the elasticity of the enclosed rubber bellows did not affect the torque.

The purpose of this study is to develop a holding device which has versatility and human friendliness. A conventional granular jamming gripper has some technical issues such as difficulty in holding small objects, gravitational influence on granular. Therefore, we develop the hybrid jamming holding device composed of granular and porous materials and a thin sheet. This hybrid structure can improve holding performance by compensating for a mutual weaknesses. In addition, a bending actuator is installed on the holding device to grasp an object. In this paper, we describe principle and structure of the device and then the grasping performance is confirmed experimentally.

This paper describes a new type of pneumatic artificial muscle, the double acting pneumatic artificial muscle. The double acting pneumatic artificial muscle has been developed based on the rubber-less artificial muscle. It is expected to reduce hysteresis characteristics of the rubber less artificial muscle. In this paper, we explain about the basic structure of the double acting pneumatic artificial muscle. The hysteresis characteristics in the relationship between pressure and displacement is examined experimentally, and we confirmed the hysteresis can be reduced by adjusting the pressure of this pneumatic artificial muscle appropriately.

The omnidirectional motion can move any direction without changing the direction of its own. The authors have developed Wave-wheel mechanism and Screw-type differential rotating mechanism. Also, the stepping performance were improved by extending the ground point from point to line. In addition, we have evaluated the motion characteristics of one ciliated wheel mechanism. In this paper, we examined the composition for vehiclization. In the vehicle configuration, the minimum necessary two units are arranged in parallel. We conducted experiments using actual mechanisms and confirmed that the inventive principle is also effective for vehicle construction, omnidirectional motion.

We developed two types of non-wearing power assist devices that allow the squat method. One is a pneumatic power assist device that uses a pneumatic actuator, the other is a passive power assist device that uses only springs. The support effect and muscle burden over time were evaluated using EMG and floor reaction force through lifting motion support experiments for these devices. From the experiment result, it was confirmed that the muscle burden on the lower body of the subject were significantly reduced, which prove the effect on one-side support using both non-wearing assist device.

This paper proposes a robust imitation learning framework with a Coarse2Fine policy to achieve repetitive long-horizon robot tasks. We formulate the framework that employs policy robustification by disturbance injection and the Coarse2Fine policy, dividing a policy into coarse policies and fine policies to reduce the amount of training data and the subsequent demonstration burden. We verify the effectiveness of the proposed method with repetitive long-horizon tasks, and the proposed method achieves the best performance to the conventional methods.

Our research group has been developing a "Joint Independent Forward/Inverse Kinematics" calculation method that can easily set up kinematics calculations for robots with various structures. Based on the proposed method, we have developed a kinematics calculation tool that runs on a Linux-based OS to easily utilize the results of this research. The kinematics calculation tool can be used by manipulating the robot structure data on the shared memory by commands. Therefore, commands can be invoked from various programming languages, and user interfaces can be created according to the purpose. In this report, we describe the details of the kinematics calculation tool that can be executed on a Linux-based OS and an actual robot developed on the premise of using this tool.

In this study, we propose a reinforcement learning method for motion instruction that aims to reduce the burden of learning on the actual robot. In this method, we construct a virtual environment for learning on a simulator, and the motion is instructed on the simulator as pre-training. The actual robot learns the motion based on the results of the pre-training. By utilizing the pre-training results, we aim to shorten the time required for learning on the actual robot and reduce the burden on the actual robot. As an example problem, we apply the proposed method to the picking motion using a vacuum suction cup of the desktop robotic arm. Through the application to the example problem, the feasibility and validity of the proposed method are basically investigated.

This paper propose a method of redesigning the reward function by setting a score for the path and changing the weights of the features. That make up the path score from the inverse reinforcement learning. It is explained in the redesign to improve the problem of avoidance accuracy of the robot.

We have developed an unmanned aerial vehicle (UAV) equipped with independently driven flapping wings. It can change its flight attitude seamlessly between hovering and level flight. The main method employed for controlling its flight attitude is PID control with an assistance of learning P gain parameter using Deep Q Network (DQN). The program is specifically designed for controlling the yaw amplitude of the airframe. The result shows that the PID-DQN hybrid method is much more effective to the ongoing attitude change than the simple PID control without DQN.

This study introduces a training system of action models for mobile robot navigation based on reinforcement learning with monocular camera input. The authors have developed a synthesis system that includes automatically building simulation environments and training of models. The purpose of this paper is to obtain a general-purpose action model by quantitatively verifying the effects of the proposed system. The simulation result shows that the trained model with the proposed system provides stable navigation performance.

Reinforcement Learning (RL) is a trial and error process where a robot interacts with its environment using a stochastic policy. For example, it is realized by adding Gaussian noise to a deterministic policy. Therefore, a straightforward application of RL to robot control tasks is often problematic because the stochastic policy does not produce smooth behaviors. We propose model-based reinforcement learning for learning a deterministic policy to overcome this issue. First, we formulate the RL algorithm with entropy regularization of the model. In this formulation, the robot explores the environment based on the simulated environmental uncertainty. We utilize the stochastic value gradient method for this formulation. Then, we derive a model learning algorithm inspired by density ratio estimation. Our proposed method is evaluated on three benchmark tasks provided by the DeepMind Control Suite, and the experimental results show that our method can produce smooth behaviors and outperform the other baselines.