This paper discusses the improvement of a gait in prosthetic leg walk by improvement of a toe part of the prosthetic foot. The toe part of the prosthetic foot was improved so that it can bend more than before. Then, the walk analysis by video, measurement of the reaction force from the floor by pressure sensor, and measurement of the burden added to the muscle of thigh by surface electromyogram were performed during the mimic prosthetic leg walk by a healthy person, and the prosthetic leg walk before and after improvement was compared. The following results were obtained after the improvement. (1) Swing of the walker's upper half of the body was reduced. (2) The impelling force from the floor increased, since the variation in pressure distribution of the sole was decreased. (3) The burden added to the muscle of thigh was also decreased.

View full abstract

In this paper, an electric wheelchair control system with Brain-Computer-Interface (BCI) is discussed for a disabled person who cannot operate joystick by his/her hands. A biosignal acquisition BCI headset, Emotiv EPOC which can detect electroencephalogram by 14 electrodes and head movement by two gyroscope sensors, is attached to the user's head, and expression of user's face and movement of user's head are distinguished. Then, control command to the electric wheelchair is assigned to each user's motion. Thus, using the motorized control device of the joystick developed in our previous study, the electric wheelchair can be controlled by user's face expression and head movement, without using user's hands. In order to evaluate the developed system, experiment which investigates accuracy of recognition for the control commands and slaloming experiment were carried out. The results showed the effectiveness of the proposed system, while some improvements to enhance the operability are required.

View full abstract

Home oxygen therapy (HOT) patients always have to carry an oxygen equipment in order to reduce shortness of breathing. We are developing a practical robotic follower carrying their oxygen equipment instead of patients. In this paper, we bring the prototype model into public transportation such as bus, train and bullet train to evaluate its size. We confirmed that the 6-th prototype model fairly fits these public transportations and can refrain from interfering.

View full abstract

In this paper, feedback of temperature perception for myoelectric prosthetic hand users is discussed. First, temperature is detected by a temperature sensor. However, since response of the sensor is late, temperature prediction was performed using the rate of the temperature variation. As a result, the temperature can be estimated in 2 or 3 seconds. Second, using the Peltier element, a feedback device of temperature perception (FB device) to be attached to the user's upper arm was developed. Third, the difference of feeling temperature between fingertip and upper arm was investigated. Then, when the user touches an object, temperature at fingertip is estimated by the temperature prediction algorithm, and the corresponding feeling temperature at upper arm is generated by the FB device. Thus, the user can feel the temperature of the object at upper arm. Finally, distinction experiment between two different temperatures was carried out to verify an effectiveness of the developed FB device.

View full abstract

This paper designs and evaluates a pair of walking assistive devices based on the arc foot of a passive walking robot. We aim to enable elderly persons to walk with less muscular strength. This device is a kind of shoe that has an arc-shaped sole designed in order not to occur speed loss when walking. The arc-shaped sole substitutes for the ankle joint of a human. We evaluated the effectiveness of the assistive devices using an electromyograph and a motion capture. The result shows that less use of a front thigh muscular is necessary.

View full abstract

We propose a pair of step-climbing units for a standard manual wheelchair. The target is a wheelchair user who has no disabled upper limbs and hopes independent life without caregivers. When traveling by a car, it is easy to load them because their weights are light and they are not attached to the wheelchair. These devices are used combining with a pair of mobile slopes and move on the slopes. Therefore, they produce propelling forces to the rear wheels of a wheelchair. The shape of the unit is like a wedge, and each unit comprises a DC motor and a crawler belt. We also discuss the controller using position and electric current data. Experimental results of the motion of climbing up a 30 cm step, which is corresponding to the height of two standard steps, confirmed effectiveness of the design.

View full abstract

There are people who can't transfer on their own and needs help for the move from a sitting posture to other sitting posture. However, many of them want to transfer without assistants. Therefore, we developed a robot that assists them to transfer on their own. A conventional robot was not able to be used by the people who could not sit up to sitting position from supine position. Therefore, we suggest self transfer assist robot which can assist sitting up movement to realize lifting of the one. In this investigation, we produce body hold part to enable holding of the person of cover assistance at the time of lifting assistance and sitting up assistance experimentally on developing the new machine.

View full abstract

In aging society, there is a growing importance of a walker for an aged person helping with the walk. However, it is hard for a walker to ascend a step. In this paper, we developed a new mechanism using a sled-like shape for riding over a step. We regarded the angle against a step as an important parameter and focused attention on it. Based on a simulation of riding over a step and the theoretical value of the impact force, an optimal angle of the sled was obtained. We will take this result into account and develop the actual machine.

View full abstract

Electrical shoulder disarticulation and Electrical above elbow prosthesis is applied to patients with high upper limb amputation. This kind of upper extremity prosthesis has danger of contacting with user's face, because it has high DOF mechanism and large working area. To ensure safety, the safety system should monitor position of user's face and prevent user's face from contacting with hand. As methods of estimating position of user's face, there are Stereo vision and TOF (Time of Flight). However, we cannot install these system in Electrical upper extremity prosthesis, because it is unsuitable for complex movement of upper extremity prosthesis. Measuring image by a single camera can estimates distance to face by counting extracted pixel, however, there are some limitations to its use in ensuring safety, mainly that it cannot estimates distance at closely position. To cope with these limitation, in this paper, we report the distance estimation method using low resolution infrared array sensor that can obtains the image of temperature distribution and be installed at significantly low cost and result of accuracy evaluation.

View full abstract

This paper proposes a novel tongue-operated joystick device for severely disabled persons. The human tongue has superior movement and tactile sensations. By providing force information during the device's operation, user can skillfully control life supporting equipment, such as electric powered wheelchairs, robotic manipulators for meals through interaction with this equipment. Therefore, we designed a joystick mechanism that enables not only the generation of control commands for the equipment but also the adjustment of the reaction force during its operation. The adjustment of the reaction force enables the user to know if there is any interference between the equipment and the environment and to improve task performance. In our proposed mechanism, the elastic pins surround the tip of the joystick and restrict its movement by this contact. The reaction force is controlled easily by adjusting the support length of the elastic pins. This paper describes the detailed mechanism and its basic performance through several experiments.

View full abstract

In recent years, more than half of the people in Japan are having low back pain. Low back pain has become a social problem in Japan. Therefore, we developed a power assist suit of the endoskeleton type that it can exert a large force to assist the human body works. We have selected the pneumatic artificial muscle as an actuator. Also, amplification mechanism is attached in the waist to efficiently transmit the power of the artificial muscle. And, we have evaluated using the VAS and EMG to evaluate the effectiveness of the power assist suit. In addition, we were confirmed objective usefulness and subjective usefulness from their results.

View full abstract

It is one of the significant problems for hospital patients to transfer from the bed to a wheelchair. This study focused on the transfer motion from the bed and developed assist system of transfer motion for the bed user, without lifting motion performed by the caregivers. Thin skin career sheet was inserted below the patient sitting on the bed side, and it was hanged on the cantilever chair equipped on the omnidirectional wheelchair. We propose a method to decrease the friction caused in the inserting the career sheet by changing the patients posture.

View full abstract

In this study, we investigated characteristic analysis of lower limbs exercises at the time of the foot device operation. The foot device was composed of the virtual keyboard and the virtual monitor, and it could provide users both position of the foot on the virtual keyboard and the selected character key. In the foot device operation, it was found that not only the toes but also the ankle moved in the side moving operation.

View full abstract

A lot of the power assist suits using pneumatic artificial muscles for assisting the extension of the knee have been developed. But their center of rotation is fixed. The knee joint of human has a complex movement. Bending and stretching motions of knee joint consist of rolling and sliding. Therefore, we developed joint mechanism for power assist suit on basis of motion mechanism of human knee joint and evaluate it.

View full abstract

We have developed a new Ankle foot orthosis of light weight, and control of the plantar flexion and the dorsiflexion. The orthosis has two C-shaped springs on the side and a solid damper on the back. We designed a C-Shape spring and solid damper with considering their mechanical properties and ranges of motion (ROM) of ankle. C-shape spring automatically fit to many user's because it does not have a fixed axis. We experimented and evaluated the prototype by EMG and ankle ROM. Ankle of motion and pretibial muscle of EMG decreased at initial contact, but they increase at other.

View full abstract

The purpose of the study on wearable power assist devices is to develop support system for those who require care. The power assist device is get heavy when the degree of freedom increases so as to prevent disturbing the movement of the human. In this study, the purpose is to verify the meal behavior to light the weight of power assist device by calculating the optical degree of freedom through mode analyses using singular value decomposition method. As the result of analyses, five degree of freedom which correspond to major modes of meal behavior. And to determine the degree of freedom of power assist device, verified the workspace of five degree of freedom manipulator which is the model of assist device by calculating the manipulability measure of manipulator.

View full abstract

Recently, the number of the fatal accident of elderly people has been increasing in Japan. Development of wearable assist systems for elderly people is studied as one of countermeasure. But, size of assist machines grows bigger and weight gets heavier, if its degree of freedom increases so as to prevent obstructing the movement of human. The purpose of this study is to extract the major modes of sit-to-stand movement using handrail for design small and lightweight assist machine by assign degree of freedom on the optimal positions. Measurement of sit-to-stand movement was used motion capture system. The major modes were extracted by the mode analysis. As the result of analysis, the coxa and knee flexural extension, adduction and abduction ware identified as major modes of human sit-to-stand movement. These motion ware inferred from preceding studies that it indicate the movement to lift a body and to stabilize posture.

View full abstract

There are people who can't transfer on their own and needs help for the move from a sitting posture to other sitting posture. However, many of them want to transfer without assistants. Therefore, we developed a robot that assists them to transfer on their own. The convention robot was not able to be used by the people who cannot hold sitting position on their own. For larger users, we thought that to realize the transfer of bedridden people. In this investigation, we attempt to add the mechanism for sitting up assistance to sitting position at a bed end from supine position.

View full abstract

In this study, we focus on the standing-up motion assist for elderly people, and we are developing a standing-up motion support system, because the support of stand-up motion is the most physically burdened task for caregivers. The system is designed to be used easily and can carry out minimum necessary support for elderly people by moving the handle. Support system consists of two electric sliders and a handle with 6-axis force torque sensor. In this paper, we aim to get a suitable position of the handle at user's seating and standing-up conditions. We measured the position of the handle, shape of user's body, and evaluated the relationship between them. Then, we found that the suitable position of the handle is determined based on the "arm's length" when sitting, and the "stature" when standing-up.

View full abstract

Home Oxygen Therapy (HOT) patients need portable oxygen equipment at the time of going out. However, the currently-used oxygen equipment is heavy enough to discourage HOT patients from their going out. Therefore we try to develop the power assist robotic cart which supports HOT patients' going out. Especially we try to develop the carry-cart type mobile robot which supports HOT patients by applying inverted pendulum control and force control. In this paper, we try to verify the effectiveness of the hybrid control system via computer simulation by Open Dynamics Engine.

View full abstract

In this paper, we propose a driving assist system for electric wheelchair. In several countries as the rate of senior citizens increases, the number of electric wheelchair users and the accidents by the senior becomes larger. To decrease these numbers, the wheelchair can avoid or stop before the confliction with the obstacles, is needed. Using the pointcloud from tilted 2D LiDAR, we construct the environment map around own position and generate the practicable paths. Considering the driver's command, safety, and smoothness, the path is selected and followed finally. To prove the validity of our method, we ride on the wheelchair and drive the several environment in our campus.

View full abstract

In swarm robotics, one of the important purposes to be required is to deploy robots to desirable formation. For the swarm robots to make pre-defined formation, it is necessary for robots to exchange information (e.g. positional information). But communications among robots are not always guaranteed because of the existence of obstacles, signal jamming, and so on. If communication loss has occurred, we need to estimate lost information by some ways. There are a lot of estimation methods from simple ways (e.g. zero-order hold) to theoretically accurate ways like Kalman filter. In swarm robotics, because robots have relatively poor computation ability, the computational cost of estimation method is desirable to be low. So, we take the three simplest estimation methods, zero-order hold, linear extrapolation, quadratic extrapolation as example, and validate their effectiveness. Moreover, we propose novel extrapolating method using exponential function and show its effectiveness through computational simulations.

View full abstract

This paper treats a swarm robots system that constructs a given structure by collecting distributed materials in the field. In our proposal, the robots with simple algorithm and the materials with simple rules construct the structure cooperatively. Efficiency of our proposal is confirmed by computer simulation and small scale experiment by real robot system. We also try to enhance its adaptability in which the thickness of the wall is reinforced by the strength of external stress.

View full abstract

In this paper, we developed a multi-agent model in order to describe trail-laying behaviors of ants. Agents in our model don't non-linearly response to chemical pheromones directly. Rather, they change their detection regions depending on the differences between right and left pheromone concentrations (|R-L|) when total amount of pheromone concentrations (R+L) beyond threshold values. If these values (R+L) are under threshold values, agents react to moving-directions of others locally detected by them instead of reacting to chemical pheromones. This switching model (reacting to either moving directions of others or chemical pheromones) appears to show non-deviated responses compared with following proposed model. In order to maintain the balances between exploration and exploitation, here we show that agents must produce deviations regarding to directions i.e. detection regions based on the relationship between moving directions of others and chemical pheromones.

View full abstract

The goal of this paper is to develop a distributed localization of mobile robots in an agricultural field. Ouragricultural robots include a GPS sensor, which does not give us an enough accuracy of position. However, the robotscan communicate each other for information sharing. Therefore, we can estimate a distance between robots by thestrength of radio signal in communication. For the scalability of a robotic system, we develop a localization methodin a distributed manner such as each of the robots estimates positions of itself and neighbors, and shares and updatesthem with other robots with Kalman fiter. We call it distributed Kalman filter by mutual observation. The developeddistributed localization method is verified with numerical experiments.

View full abstract

There are various creatures groups in the natural world. For example, they are fished, are birds, are ants, and are bees. Among them, birds and fishes organize a group by the local interaction. In such group, a leader does not exist. And the number of individuals organizing a group does not have any limit. So these group can organize massive herd easily. This feature is useful to organize swarm robots systems in the place where it is difficult for a human being to approach. The authors thought that form of swarm can measure the features of three-dimensional space. Therefore, they authors used the algorithm known as "Boid", and organized swarm of agents. In this paper, the authors have verify if a swarm can measure the features of three-dimensional space.

View full abstract

One approach in swarm robotics is homogeneous system which is embedded with sensing, computing, mobile and communication components. In this study, a target detection problem, which is one of navigation problems, was employed. Once a robot detects a target, robots immediately communicate with a base station via intermediate relay robots due to the multi-hop transmission of wireless communication. Therefore, it is important to take connectivity of the network into account. This study investigates the performances of the two kinds of random walk algorithm in navigation while loosely ensuring connectivity of the robotic network based on our previous study.

View full abstract

This paper proposes a new response threshold model to implement an autonomous specialization for multi-robot systems. The response threshold model is a computation model of the sensitivity of ants to external stimuli. Each ant has the different sensitivity to external stimuli. Thus, it is considered that ants can specialize either of one of the active ants or the non-active ants autonomously. The conventional model uses the ratio of active ants in a colony as external stimuli. However, ants cannot perceive all information of the colony whole. Therefore, the proposed response threshold model uses the ratio of active ants that an ant came across during a certain term as external stimuli. In this paper, we investigate the efficiency of the proposed response threshold model through a collecting foods problem in simulation experimentations.

View full abstract

In this paper we evaluate Boid-based flocking behavior of a real robotics swarm in various input-output specifications. Boids model is assumed that each individuals can obtain information of other individual. However, robots in a real environment required many sensors and complicated sensor to obtain information of other robots. On the other hand there is research using binary sensor for aggregation. So we simplify input and output information for flocking. We compare performs simplified vision to recognize other robots and compare proportional control and simple control. Generated flocking behavior is evaluated quantitatively.

View full abstract

This paper proposes a behavior model for swarm robotic systems that can be externally guided, with agreement on their heading direction with other robots. Generation of herd behavior in the real environment is not only simple tasks successfully, such as moving to straight behavior and destination. To achieve the reciprocating between two points is a more complex task, we added the approach of changing the behavior of each robot stochastically during the experiment to the Boid. Each robot to update the probability learning based on local information around. By conducting physical experiments, the effectiveness of the proposed method is evaluated quantitatively.

View full abstract

Even if the ability of one individual is low such as an insect, a fish and a bird, are often takes orderly action by forming the swarm. This is called swarm intelligence. There is swarm intelligence. The purposes of this research are realization of which is based on group of nature with enhancing of crowd action to three dimensions by using flying robots, and analyze it. The habit that a fish forms a group and lives by becoming a group was inquired using the airship robot. And by increasing the number of robots airship, Behavior such as more fish swarm I thought to be able to reproduce. Extension into the three dimensions of group action was realized by making group action of fish into a norm for the airship.

View full abstract

In decision making process for games with perfect information, some well-known techniques, e.g., mini-max algorithms, are useful. However those algorithms are difficult to apply for the games with imperfect information. In this study, the UCT search is applied to make a decision for the Hanafuda game that is a Japanese traditional card game. Hanafuda game is also one of typical games with imperfect information. We attempt to acquire a strategy of this game by using UCT search. Some computational experiments are carried out to verify usefulness of our approach. Obtained experimental results illustrate the effectiveness of UCT search.

View full abstract

In this study, we developed Android-based controller for humanoid robot. In conventional hobby robots, a specialized controller with low-performance microcomputer was mounted, so it is not suitable for high-load computation, such as image processing or real time gait generation. Then, we attempt to replace it with a Android terminal. Android terminal are easily obtainable, lower cost, highly performance, and quick release cycle. We conducted some experiments to walk on a slope using an acceleration sensor on Android terminal by solving real time inverse kinematics for leg motion.

View full abstract

In multi-agent reinforcement learning where the whole system receives one reward, it is necessary for effective learning to distribute reward for each agent appropriately. In this paper, we define a value called "confidence factor", and suggest a novel reward distribution method using this value. In this method, each agent calculates own confidence factor for the success of the task using its peripheral information. Then, the contribution factor of each agent is estimated and the reward is distributed between agents. Using the peripheral information of each agent, effective reduction of the observation cost of state-space and appropriate reward distribution becomes possible. By simulation tasks, we empirically show the effectiveness of the reward distribution of proposed method.

View full abstract

Currently, the development of a variety of robot is in progress, but the research that a robot coordinated action does not advance. In this study, we tried to develop a smart control mechanism of appropriate behaviors of swarm robots for multiple cooperative tasks. Given cooperative task is a 3-Dimensional balancing task. Swarm robots system has to satisfy both tasks simultaneously. For this purpose, swarm robots have behavior controller constructed by artificial neural networks (ANN). The ANN controllers output control signals of motors from sensor values of each robot. Weight values of ANN controller are optimized by particle swarm optimization technique which is one of useful meta heuristics. Based on the proposed approach, computer simulations are carried out. Obtained experimental results illustrate effectiveness of our approach.

View full abstract

This paper describes an assimilation method of knowledge obtained by reinforcement learning agents. In recent years, multi-agent robot system (MARS) utilizing reinforcement learning and its knowledge sharing methodology have been deployed in real-world situations. For example, knowledge assimilation methods such as simple-sum are proposed for merging the knowledge obtained from distributed learning agents in different tasks. Those methodologies, however, have not been fully discussed systematically. Furthermore, existing researches have only been dealing with the knowledge which is obtained from individual tasks. To realize a knowledge assimilation method with various knowledge of agents in multi-agent reinforcement learning systems, we focused on a cloud computing resources. In this paper, we propose a knowledge assimilation method utilizing a cloud computing resource, and we confirmed the effectiveness of knowledge assimilation method in a shortest path problem.

View full abstract

This paper describes an influence of heterogeneity for a transfer learning in heterogeneous multiple robots. A multi-agent robot system using reinforcement learning (Multi-agent reinforcement learning: MARL) is effective methodology to obtain efficient behaviors autonomously in dynamic environment. However, the MARL has a problem such as prolongation in learning time. The existing research indicated effectiveness of transfer learning for reducing of learning time in reinforcement learning. Transfer learning is a framework of reuse of knowledge which is obtained by reinforcement learning. In our prior research, we proposed and applied the parameter, transfer rate, to the transfer learning method for heterogeneous agents. The parameter adjusts the degree of reuse of past knowledge toward a newly obtained knowledge in a new task. However, transferability is depending on heterogeneity factors such as difficulty of tasks and agents' functions. In this paper, we investigate the effectiveness of transfer rate with heterogeneous agents and environments by conducting a computer simulation.

View full abstract

This study aims the development of methodology to deform ultra redundant robots according to commanded operations and generate motions of those. The important points in the methodology are the modularize of the forward/inverse kinematics calculations every joints for forming robots, and the multi-point positioning of ultra redundant robots. Because inverse kinematics calculation modules calculate own angle without relationships of a whole structure of robot, multi-points positioning problem become simple problem. The forward kinematics calculation modules are connected each other corresponding to structure of robots. These connections are to making flow of calculation and it can be branch structures calculations. This report describes details of the modularize of the forward/inverse kinematics, the method of the deformation of robots. Finally, some simulations are indicated for confirmation of effectiveness of the proposed method.

View full abstract

In real world, modification and generation of locomotion are important for autonomous moving robots as well as animals. Many studies show that central pattern generators (CPGs) allow robots to produce robust periodic motions. However, generating substantially new locomotor behaviors is still difficult because methods such as CPGs require parameter tuning in advance. In this paper, we show that maximization of mutual information for sensory time series allows robot system to generate locomotor behaviors. Using simulation we found that quadruped and snake robots can generate several locomotor patterns with several environmental setups although with no prior knowledge for their bodies and environments. Furthermore, both in computer simulations and robot experiments we show that maximization of mutual information enables soft-body robot to generate locomotor behaviors in a self-organized manner. Our results suggest that methods focusing on sensorimotor information may realize autonomous moving robots which can keep moving when unpredictable changes in their bodies and environments occur.

View full abstract

To design a 3D biped robot, we should consider both of sagittal and lateral planes. Since the 3D biped robot involves two swings in the lateral and sagittal planes, passive dynamic walker in the lateral plane for synchronizing the swinging in the two planes and walks stably in three dimensions. -First, we designed a lateral model that can change radius of curvature of a foot. By changing a cycle of walking steps in the lateral can be controlled. Second, we developed a method of synchronizing two cycles of walking steps in the two planes by changing the radius of the foot. We performed numerical experiment for the model.

View full abstract

Use is expected of redundant manipulator that combing through rubble after an earthquake and a tsunami etc. It is necessary that controlling embodiment or the feature and joint position of the redundant manipulator, operating the redundant manipulator applied to the environment and the work situation. But in some cases, it is difficult calculating exact solution for embodiment control of redundant manipulator. This paper deals with a kinematics resolution framework for robots controller to perform multi target positioning. The framework consists of metaphysical subsystems, corresponding to joint and link, and their connection rules. The localized forward and inverse kinematics calculations are embedded in the subsystems and kinematics models are composed of a set of subsystems assembled by their connection rules. The framework is applied to case studies of the forward and the inverse kinematics problems which are multi target positioning of redundant manipulator.

View full abstract

Perceptual Crossing experiment has been proposed to analyse the human social interaction with low dimensional sensation and motion. The previous study shows that human subjects can distinguish the partner with the distracted moving objects by coordinated behaviors established in two. In order to investigate the coordinated behavior established by human subjects, we design different types of AI programs and test how those programs can interact with human subjects. Our result shows that the turn-taking type AI program is better to reproduce human social interaction.

View full abstract

The Job-shop Scheduling Problem is a classical, but yet modern problem. Because it belongs to the NP-hard problem and it is almost impossible to find out the exact solution, many heuristic methods have been proposed. Recently, Immune Algorithm (IA) and Genetic Algorithm (GA) so on have been proposed as a powerful tool to solve the problem. The optimal schedule becomes very important from a viewpoint of cost reduction in the manufacturing industry. This study optimizes the job-shop scheduling with multiple purpose functions by using evolutional computation. Numerical experiments verify that the evolutional computation obtains (quasi-) optimal schedule with multiple purpose functions.

View full abstract

We have modeled autonomous mobile robots by applying the physical engine (PhysX) to a virtual environment. This study aims at acquiring an autonomous and cooperative behavior for autonomous mobile robots, which moves from a start place to a destination place. Autonomous mobile robots are controlled by Artificial Neural Network (ANN), which is evolved by learning proper actions by use of Genetic Algorithm (GA). Simulation results show that autonomous mobile robots acquire autonomous and cooperative behaviors under the Virtual physical environment of three dimensions.

View full abstract

This study aims to propose an artificial neural network based on network topology and learning rule (Spike Timing Dependent Plasticity) of cerebral cortex. Numerical experiments showed that proposed neural network has capability to organize functional area that belongs each input signal.

View full abstract

This paper describes a behavior acquisition of virtual robots by a gradual learning. The gradual learning is a method in which initial states of simulation for evaluation is changing in evolution progress. Motion of virtual robot is calculated by the physical engine PhysX, and it is controlled by artificial neural network (ANN). Parameters of an ANN are optimized by particle swarm optimization (PSO) so that a virtual robot follows the given target. Experimental results show that the gradual learning is better than a general learning, and that a random initialization of parameters in the middle of evolution leads to better evaluation. The best motion given by the gradual learning is composed of some basic motions, which are controlled by the ANNs used for change of initial states.

View full abstract

Sound is an important information for the recognition of the environment. In recent years, the performance improvement of various recognition task by deep learning have been reported, is also expected to be applied to event detection in acoustic signal. In this paper, we discuss acoustic event detection using a deep learning.

View full abstract

The purpose of this research is to develop an automatic design method for hardware and software of Modular robot. In this paper, the genetic algorithm is applied as the design method for shape and behavior of a virtual mobile modular robot. The purpose of the design of modular robot is to track the light source. Designed modular robots are simulated in a three-dimensional virtual physics space, and evaluated by light source tracking capability. As a result, an appropriate shape and behavior for virtual robots is acquired.

View full abstract

This paper verifies the improvement of learning parameters by parameter adjustment. In this paper, we used predictive control system using Back Propagation Through Time (BPTT) introduced recurrent RBF networks (RRBFN) and Fuzzy rules. This system constructed from a prediction system and a Fuzzy control system. The prediction system predicts the state of the controlled object on time t+n. The Fuzzy control system controls the controlled object based on the prediction results of the long-term prediction system. We test the proposed method under an outfielder problem in order to investigate its efficiency.

View full abstract

This paper describes a convergence estimation method on fractal dimensional analysis for reinforcement learning of an autonomous mobile robot. A multi-agent robot system (MARS) that utilizes reinforcement learning has recently been studied in real-world situations. In MARS, autonomous agents obtain behavior autonomously through multi-agent reinforcement learning (MARL). In recent years, methodology of MARL utilizing the cloud has been discussed to share the obtained knowledge among heterogeneous robots. However, knowledge sharing is ineffective while the agent is still learning. Therefore, in prior research, we developed an autonomous convergence estimation method on fractal dimensional analysis with image processing (CEFDi), and we evaluate the effectiveness of CEFDi by conducting computer simulations. Here, we develop CEFD with numerical data (CEFDn) for actual mobile robot implementation and evaluate the effectiveness of CEFDn in actual shortest-path problem.

View full abstract