Living organisms have the capability of adapting to unknown environmental situations. However, the capability of brain is finite even though environmental situations are infinite. We consider biological learning method and information processing structure makes such features. Many researches indicate living organisms use synergy, and make high dimension signal to low dimension, and vice versa. We proposed a learning method that embodies the capability of biological systems, termed "Tacit learning". However, no research indicate living organisms are learning and controlling their body by using synergy. Tacit learning is conventionally applied to the joint space, and we realized bipedal locomotion of robot. In this research, we extend tacit learning to mode space decomposed by using a resonance mode of the robot. We show the robot realize high robustness using physical characteristics.

View full abstract

Recent reports revealed that in some cases the behavior in living things became probabilistic and in other cases became deterministic in order to adapt to the complex environment. This mechanism is called as "Yuragi", which means fluctuation in Japanese, and can be represented as a stochastic differential equation and referred to as the "Yuragi" equation. The conventional "Yuragi" equation makes use of both "task achievements" as an evaluation criterion and "ever-present noise" as an exploration source. Instead of the "task achievements", our "Yuragi" equation utilizes "the appropriateness of each attractor in the state space" and "damped noise according to the distance from the attractors", which offers an efficient and robust optimization algorithm. The convergence of solution of the proposed "Yuragi" equation was investigated through simple numerical calculation experiments.

View full abstract

To modify robot's motion, intuitive instruction methods such as using gestures or natural language has been proposed. However, it is necessary to define each word or gesture as a concrete amount of modification of the robot motion. We propose a motion modification method for robots by using subjective motion-feature axes. By the method, we expect a robot to modify its motion according to the intuitive and subjective words, which are not connected to physical quantity, such as "put some snap," "be cooler," and "make it a little more beautiful." Using the method, a robot could modify some motions according to the modification instruction based on a subjective motion-feature space.

View full abstract

Recent years, soccer simulation (Robocup) using a robot agent has been actively conducted. In this research, we are focusing on basketball shooting motion. Here, In order to acquire the best performance of the robot agent, we are using Reinforcement Learning with fuzzy reasoning (fuzzy Q-learning) for the shooting motion. Furthermore, we utilize Open Dynamics Engine as a physics simulation to verify the capability of our method.

View full abstract

There are many moving obstacles in RoboCup Middle Size League (MSL). A robot must be able to avoid these obstacles to prevent damage. A path planning algorithm is required. Dynamic window approach is one of the most common path planning algorithms for mobile robots. It produces a path in each control steps that is considered as robot dynamics. However, in Dynamic window approach, centrifugal force is not considered to estimate the trajectory for the robot. In this study, centrifugal force to Dynamic window approach is considered and tested by a simulation. As a result, obstacle avoiding ability without a large amount of centrifugal force is achieved compared to the existing method.

View full abstract

In late years, Japanese agriculture is aging and young people away from primary industries. For these problems, by using the robot and information communication technology, labor saving, automation, and "smart agriculture" that enables improvement of productivity is expected. Tomato is a kind of major vegetables, harvest and shipping amount is located in the upper level. However, the time required for harvesting to the total working time per 10a accounts also 30%〜40%. Also, many of the tomatoes are grown indoors, because of the high temperature and high humidity environment, labor burden greater labor-saving is desired. Studies have been made extensive aimed at commercialization of tomato harvesting robot so far. We are through the tomato robot competitions that aim to automate and practical application of the harvest of vegetables.

View full abstract

This paper describes the robot contest produce competition and its purpose on creativity education. In this competition, participants plan an original robot contest for children including the rule book, the field, the robots for demonstration and the promotion video. Then, they present the robot contest as the trial management. This competition aims to promote their creativity, especially the problem-raising ability and the problem segmentation ability, and presentation ability too. The robot contest produce competition 2014 is also reported as an embodiment.

View full abstract

RoboCup is an international join project to promote fields such as Artificial Intelligent (AI), robotics, and related field. It is an attempt to apply AI and Intelligent robotics research by providing standard problems where a wide variety of technologies can be integrated and examined. In RoboCup, soccer game is selected as a main topic of research, aiming at innovations to be applied for socially significant problems and industries in the future. We also have organized RoboCup team "Hibikino-Musashi" for Middle-Size League soccer competition. In this paper, the motion analysis and control of the ball operation at the time of ball dribbling is described.

View full abstract

We are studying on a color change of a ball image captured by a camera on a humanoid robot. In RoboCup Humanoid League, most robots detect objects such as a ball or robots by color-based image processing. Object colors are categorized using a pre-calibrated look-up table. However, color of the ball is changed according to a distance from the camera to the ball. Therefore, we conducted to clarify the mechanism of the color change of the ball. Through experiments, we found that one of the reasons is related to a low pass filter in the camera. We propose a model of the color change and evaluate its effectiveness.

View full abstract

In the RoboCup soccer medium-size league, self-position estimation using the white line information is performed. In my team, takeover of software technology is difficult, and that is a problem. So, We have developed a self-position estimation program using a particle filter with MATLAB / Simulink. We shows the results of the program of precision experiments.

View full abstract

Action and strategy of the robots is sophisticated in RoboCup soccer middle sized league. Concerted action by multiple robots depends on the circumstances of the operation. Behavioral algorithm that can handle dynamic changing environments is necessary. We aim to realize the ball passing behavior in concerted action. We report on the high speeding technique of path point selection algorithm.

View full abstract

Recently, a household robot takes action according to human instructions and operation, doesn't assume action in unknown environment. Therefore, in order for such as a robot to coexist with human beings, it should be estimation, decision making and action automatically in various situations, and should have judgment ability as them. In this study, we are developing a storage position estimation system according to environmental information such as dish position, weight and so on. This paper discusses that the outline of robot system for clearing table, and the development of the kukanchi system which integrates various sensors fixed in a room.

View full abstract

In this paper, we propose the system which identifies user's activities and acquires a life log from impact noise of daily objects in Kukanchi, without using a camera to reduce pshycological load. The logged noise is analized in the frequency domain appling spectram analysis method to identify what things users are using, action discriminate from the combination of thing currently used. We devised the way to identify objects without depending on states.

View full abstract

In this paper, a novel technique for high-accuracy indoor positioning using a GPS-compatible positioning system called IMES (Indoor Messaging System) is proposed, especially aiming to apply it to robot navigation. The radio wave propagation is modeled in order to obtain the distances between IMES transmitters and a receiver. The position of the receiver is calculated based on the principle of triangulation with those distances. In addition, an Extended Kalman filter combined with the travel distance acquired from wheel encoders of a robot is used to smooth the positioning result. The result of a positioning experiment shows that the positioning accuracy more than 1 meter is achievable without giving an initial position and direction to the robot.

View full abstract

Simultaneous Localization And Mapping (SLAM) is one of the most important problems of mobile robot systems. Recently, 3D distance sensor is expected for 3D SLAM. However, It is difficult to build the 3D environmental map in real time because a distance dataset from 3D distance sensor has many position data and it is difficult to search the corresponding points from the dataset in real time. In our research, we combine the 2D environmental maps that are divided into different height for realizing the real-time 3D SLAM. In this paper, we propose a real-time 2D SLAM based on an Evolution Strategy by using 3D distance sensor. The experimental result shows the effectiveness of our proposed method.

View full abstract

In recent years, going out opportunities for the elderly is reduced, bedridden and homebound and have been a problem. This background is the lack of movement means of the elderly. In recent years, characteristics of the elderly, is that the prevalence of smart devices, such as Internet usage and smart phones, tablet computers is increasing. In this study, we have developed a smart device linked mobility scooter in order to support the day-to-day life of the elderly with a focus on smart device. And to request the analysis of functionality required in the elderly is used in this system to implement some of these functions. Also performs fuzzy control for acceleration and deceleration of the mobility scooter with a simplified fuzzy reasoning, it is possible to suppress the acceleration applied to the elderly during traveling, and improved safety.

View full abstract

In this paper, we describe a humanoid robot "Accelite" developed by our Team CIT Brains for the RoboCup soccer humanoid KidSize league. We have been participating in the Humanoid League for eight years. Two years ago, we redesigned the system to put a large weight on maintainability and usability. In RoboCup 2014, we won the first prizes of 4on4 soccer and technical challenge. Consequently, we were awarded the Louis Vuitton Humanoid Cup. The system we developed has high mobility, well-designed control system, position estimation by a monocular camera, user-friendly interface and a simulator. The robot can walk speedily and robustly. It also has a feedback system with a gyro sensor to prevent falls. It detects positions of landmarks by color-based image processing. A particle filter is employed to localize the robot in the soccer field fusing the motion model and landmark observation.

View full abstract

This research is aimed at the realization of system for coordinated action in response to a variety of field conditions in RoboCup middle-size league. Therefore, to estimate the path position depending on the situation by using a particle filter. Also, to realize the path by selecting the success rate of a path is high robot in ally of the robot. Simulation results of this approach, take a sufficient distance and opponent robots in performing the path, and can select a valid path partner, allowing the path depending on the situation.

View full abstract

In recent years, the number of systems that observe indoor space with a lot of sensors in real environments has been increasing. We propose a sensor-monitor network infrastructure, which enables any devices, modules, and robots on the network to share any sensors' information. Thanks to this infrastructure, the Kinect V2 server distributes sensor information to them, and the visual server integrates remote sensing data from all places and distributes them to multiple monitors on the network. On this infrastructure, for example, an indoor logging system is able to detect and interpret the people and objects related events by integrating information of multiple sensors at all places.

View full abstract

We aim to increase the density of spatial memory which is an interface to achieve storage of and access information in intelligent spaces. To realize this, we propose a hierarchical structure of the spatial memory. To build the hierarchical structure, we define the global spatial memory and the local spatial memory. The global spatial memory consists of SKT (Spatial-Knowledge-Tag) which is defined in a global coordinate system. The local spatial memory is defined as local-SKTs connected with a global SKT. A local-SKT is defined in a coordinate system of the head-mounted display. To interact with and to operate the hierarchical spatial memory intuitively, we use see-through head-mounted display and Leap motion.

View full abstract

In recent years, automatic visual monitoring system are required for security reason and the research and development of such systems are hot topics in the application field of computer vision. Amount of automatic visual monitoring system's log data increases and it has risen that they needs to search for required data. We propose a query interface for automatic logging system of person-object communication based on detecting human behaviors that occurs in a room. The detected and interpreted events are recorded in a database. This query interface that the user and the system interactively search recorded events for required scene on the browser. In this paper, we describe the method of implementing query interface.

View full abstract

In a system that can detect and record the changes that occur in the monitoring room, we focus on the people behavior and the object relevance of the people, detected and recorded the changes that occur in the room, propose a method of detection that object move by people. System can detect bring in, bring out and move object, hide motion and drink motion by people as event. And, detect the movement of the object and recognizes the object moving before and after as the same object, collects various appearance of the object, and performs detection of the operation for the object person.

View full abstract

A smart electric wheelchair for use as a personal mobility tool (PMT) has been proposed. To reduce the load on the user, the PMT is integrated with autonomous mobile robot navigation and human inputs. In addition, an obstacle avoidance function must be helpful for the user. Comfortable and relief movement is required to the obstacle avoidance of the PMT for a user and people around as well as safety and efficiency of the movement. Therefore this paper aims to realize the obstacle avoidance that does not get the user nerves and discomfort. We use the virtual impedance method as an obstacle avoidance method under dynamic environments. An experiment is conducted to evaluate the obstacle avoidance movement of the PMT from psychological affect point of view when different types of movement characteristics are set.

View full abstract

Pre-calibration of camera network system consisting of distributed camera sensors (i.e., determining the absolute poses of each camera) is an essential task to construct an intelligent space. This research deals with automatic calibration method of external parameters of distributed cameras by utilizing easily obtainable grid map information of the environment as prior information. The map information help seek a global minimum solution (i.e., camera parameters) within the objective function containing many local minimums. We evaluate the proposed method with a mobile robot in a wireless camera network.

View full abstract

Increase of CO_2 and PM2.5 concentration are serious problems. The technique which measures the gas concentration in the environment is very important. In addition, today, people who require care are five million in Japan. Therefore, the technique which measures the gas concentration in room can help the people require care. In this study, small gas sensor module which enables to communicate in wireless communication is developed. In this paper, we attempted to evaluate a characteristic of solid-state electrolyte type CO_2 sensor.

View full abstract

Indication of place where the things are stored is needed to dictate the tidy things up to robot. If robot has commonsense knowledge about place where the things are stored, robot can tidy things up without indication of place where the things should be stored. I research the system to build commonsense knowledge for tidy up robotic service by using object trajectory log.

View full abstract

Parallel wire suspended system has a large weight capacity and wide range of movement as a positioning mechanism for heavy objects. In many researches of parallel wire suspended system, a sag of wire has been ignored. However, it seems that using the model that taking into account a sag of wire improves accuracy of the position control of parallel wire suspended system. In this paper, a calculation method of kinematics and stiffness takes into consideration a sag of wire is proposed. Using this method, it is shown that light weight of suspended object and high linear density of wire make large displacement by a sag and reduce the stiffness of the system.

View full abstract

The purpose of this study is to develop the spatial manipulation system by using parallel wires that are driven by the stepping motor and pulleys. This mechanism is enable of performing the operation of micro tool such the injection or other tools. In this paper, the concept and the design as well as the experimental performance are discussed and the positioning resolution of 7 micrometer has been achieved over 60mm travel.

View full abstract

Self-contained pneumatic driven musculoskeletal robots for dynamic motions are often equipped with ON/OFF valves for weight reduction. It is necessary to switch valves frequently since a discontinuous control system with ON/OFF valves causes overshooting. Frequent switching leads discarding kinetic and elastic energy. One-shot valves operation to realize a target motion is desired. Target of this research is to design a musculoskeletal structure not to discard energy in a target motion such as using maximum performance of a robot. We propose a design method of pneumatic driven musculoskeletal robots to realize a target dynamic motion by one-shot valves operation by designed variable moment arms of a musculoskeletal structure.

View full abstract

The purpose of this research is to analyze the role of walking robot's muscle. Computational model of 1-DOF leg robot with musculoskeletal system was derived. The robot's leg was moved by antagonism of two wires imitating muscles. Equation of motion for the system was obtained. Control simulation of the robot was performed based on the equation of motion. The simulation was executed on the program developed for this research. On the simulation, the robot was controlled to a desire position by torques of motors to pull wires imitating muscles.

View full abstract

This paper presents the under-actuated robot hand which is driven by wire. A new mechanism using a spring is loaded into this hand. This hand is using the finger with the system to switch two motions, joint synchronous and adaptable motion passively. In order to adapt to objects with curved surfaces, the spring base mechanism was provided such that the joint of finger root can passible rotate and the hand can adapt to the object shape. The hand can get the suitable holding form according to the object by that mechanism. The hand was mainly made by ABS and was light weight. It can reduce damage due to collision.

View full abstract

We developed a tele-operation system for a crane. The crane consists of a large actuator that transports suspended objects and a small actuator that is used as a regulator and feedback-controlled by a wire angle sensor. The operation system can convey the suspended objects safely and efficiently by moving a crane with input signals from screen click interface. However, the suspended object collides with obstacles when dynamic obstacles exist on a conveyance course. Therefore, we propose an algorithm to generate the conveyance course that can avoid dynamic obstacles automatically. The algorithm enables the suspended objects to avoid an obstacle by changing the moving velocity of the suspended objects in the two-dimensional direction that is perpendicular to the gravity direction. We verified the effectiveness of the algorithm through a conveyance experiment.

View full abstract

This report describes the design of a haptic device for the skill assist system which realizes high performance human operations. We adopted a wire parallel mechanism as this haptic device. This haptic device must not interfere with not only human but also operating objects. There is a high possibility of interference of ordinary parallel wire mechanisms With an extrusion mechanism. And to realize high transparency and large force, the new wire driving mechanism was proposed. It drives wire without reducer. But it has a driving puller and an idler pulley on which wire are winded many times. The prototype of the haptic device was designed and fabricated

View full abstract

Synthetic fiber rope which has lightweight, high tensile strength and flexibility is receiving a lot of attention as the replacement for stainless steel wire rope. This study is clarifying various physical characteristics of synthetic fiber ropes to design tendon-driven mechanism and this paper take note of durability of repeated bending. We performed an experiment with two synthetic fiber ropes and one stainless steel wire rope in conformity with Japanese Industrial Standards. As a result, stainless steel wire rope and one synthetic fiber rope which has high resistance to frictional wear didn't get loss of tensile strength, however, repeated bending causes the tensile strength deterioration of the other synthetic fiber ropes which is weak in frictional wear.

View full abstract

In this paper, we propose a novel learning method to find relation from muscle-length to joint-angle and vise verse in a musculoskeletal robot, based on nonlinear bidirectional mapping between muscle-length and joint-angle. We show that mapping from joint-angle to muscle-length can be learned as linear combination of multiplications of joint-angle. This formulation enable us to extract not only muscle-jacobian but also EKF for posture identification in a musculoskeletal robot. Finally we applied the method to tendon driven musculoskeletal robot "Kenshiro" and verified its validity.

View full abstract

The purpose of this paper is to propose a gravity load compensation method for carrying heavy load with spring and a tuning mechanism of compensation force. There is a previous work that can compensate weight of the manipulator by spring, which name is MGC(Mechanical Gravity Canceller). However, it aimed at compensating fixed weight, so in case of supporting various weight transportation, it needs hand regulation of arm parameters. The proposal new mechanism can compensate any weight of a load within a limit of the setting, so it can support the work of heavy weight transportation without a driving power of a motor. As a result of experimentation, it was shown that this mechanism could exert compensation force suitable for load.

View full abstract

The deformable object such as string, cloth or paper, is soft and can change its form easily. It is difficult for a robot to manipulate deformable object, because it needs to deal with various form. The demand is growing in industry and domestic work. On manipulating deformable objects using robot manipulator, it is important to recognize their form. In this research, distance camera is used to get point cloud data of string. In addition, point chain model is adapted to describe form of string. To create point chain model from point cloud data, vector sequence that describes center line of string needs to be made. Therefore string form recognition algorithm is suggested. This algorithm estimates center line of string using surface normal vectors evaluated by normal estimation algorithm, and creates point chain model by matching method. In addition, experiment to evaluate the effectiveness of this method is conducted.

View full abstract

This paper presents scene discrimination by using a histogram based on local image features. Local feature is often used for various applications such as object recognition and navigation of mobile robots. When the robot requires the identification of some landmarks, the feature matching against a lot of interest points can be omitted in order to decide the most similar image. Therefore we obtain the histogram of binary features detected in an image. And the histogram is compared with other histograms so as to determine the visual scene from the database. Speed up of the similar image matching is enabled because keypoints matching computation that needs long processing time can be omitted by using the histogram.

View full abstract

The ambient light is changed by the factor of nature such as a weather, season and time, and also changed by the neighboring apparatuses such as an arc light of the welding apparatus. The glossy object reflects an ambient light and cause blown out highlights. Size and the position of the blown out highlights are changed by the change of the ambient light. It is difficult to use a general image processing system on the condition that an ambient light changes. Blue light was used to solve this problem. In this paper, the inspection system of the glossy sealing component is explained.

View full abstract

Structure from Motion is one of the 3D measurement methods. This technique requires only a single camera, and reconstructs the objects from pictures obtained by the camera with moving. However there is an essential problem that the scale of the objects cannot be computed by Structure from Motion. In order to solve this problem, we have proposed scale-reconstructable Structure from Motion using refraction. The refractive plate is placed in front of a camera to generate the refraction. The nomal of the refractive plate is required to be the same direction of the camera axis in the previous study. This is a limitation when the camera and the refractive plate are set. In this paper, we propose the method that the plate is not required to be placed orthogonal to the optical axis of camera. Therefore the setting of a camera and a refractive plate is more flexible than previous method.

View full abstract

Real-time full-pixel optical flow estimation is conducted for 512 x 512 images by accelerating the Lucas-Kanade method on a CPU-based high-speed vision system. Our system can remarkably expand the measurable range of the estimate flows by implementing a novel algorithm that can optimally select frame intervals and space intervals in optical flow estimation according to the estimated flow speed. Its effectiveness is verified by showing estimated motion distributions for high-speed objects moving at 100 km/h or more.

View full abstract

This paper presents the image sensor unit for rescue devices. The image sensor device equips with the sensor to measure the motion of the camera and the direction of gravity to stabilize the acquired image according posture of the camera. The advantage of the proposed image sensor unit is also explained comparing the conventional approach. The image unit described in this paper consists of intelligent camera development components and 9 axis motion sensor. The results of the estimation of the camera angle with Gyroscope, Accelerometer and Kalman filter applied to the sensor value from Gyroscope and Accelerometer are also explained.

View full abstract

In this paper, we address the challenging problem of single-view cross-season place recognition. A new approach is proposed for compact discriminative scene descriptor that facilitates coping with changes in appearance in the environment. We focus on a simple effective strategy that uses those landmarks whose appearance have not actually changed across seasons as valid landmarks. Unlike popular bag-of-words scene descriptors which rely on a library of vector quantized visual features, our descriptor is based on a library of raw image data, such as publicly available photo collections from Google StreetView, and directly mines it to discover landmarks (i.e., image patches) that effectively explain an input query/database image. Discovered landmarks are then compactly described by their pose and shape (i.e., library image ID, bounding boxes) and used as a compact discriminative scene descriptor for the input image. We collected a challenging dataset of single-view images across seasons with annotated ground truth, and evaluated the effectiveness of our scene description framework by comparing its performance to that of previous approaches.

View full abstract

Monitoring robot has the potential to provide a variety of assist in performing complex tasks. It is desirable that monitoring robot selects an adaptive ROI(region of interest) to capture working robot and peripheral object relation. However robot cannot recognize the invisible relationship between the objects like a human. In this study, we introduce four Gestalt evaluation factors of proximity, continuity, common fate and similarity, which allow the monitoring robot to select an adaptive ROI. Experimental results show the ability of ROI selection similar to human by the proposed method.

View full abstract

In the recent years, the autonomous mobile robot has become an active research in current society. The popular research are regarding the environmental map-making and object recognition. In this paper, we proposed a method to do automatic object modeling based on the observation from multi robots. We perform the data acquisition using two mobile robots attached with RGB-D camera at several different position. We combine the multi viewpoint data which we acquired using ICP algorithm and make it as a model. The reason we make the model is to perform the object recognition in the environmental map.

View full abstract

This paper presents a high-frame-rate (HFR) and high-dynamic-range (HDR) imaging system. By accelerating an HDR image processing algorithm on a GPGPU board, four 8-bit graylevel 512×512 images with different exposure times, whose intervals are 1 ms, are synthesized in real time as an 512x512 HDR image at 1000 fps. The effectiveness of our HFR-HDR vision system is demonstrated for a dynamic changing scenario, which involves both outdoor bright scenes and indoor dark scenes.

View full abstract

Visual surveillance, which relies on human motion recognition and people recognition, has received a lot of attention for its use in effective monitoring of public places. However, there is a concern of loss of privacy due to distinguishable facial information. To deal with this issue, we developed a camera system which does riot capture any facial information. In this paper we propose an abnormal-behavior detection method using privacy-protected videos taken by the proposed camera system. In the proposed method, we extract both motion-based and appearance-based features, and we combine these two methods by taking advantages of each of them. We build a database including normal and abnormal behaviors, and we show the effectiveness of the proposed method on cases from the database.

View full abstract

We study a method for unknown object detection based on impacting and keypoint tracking. In this method, a robot changes object positions by impacting to detect each of the objects individually from camera images before and after impacting. This detection is possible because keypoints of each object always move consistently by impacting, while those of the background do not move. A concave hull segmentation method called alpha-shape is used to model the objects. Picking experiments of several objects are demonstrated.

View full abstract

This paper provides a person detection system using high-speed vision. Our system is composed of the high speed camera which takes 500 image frames for a second and a PC with a GPU. In order to detect a person in the image correctly, we use the HOG (Histograms of Oriented Gradients) which is not affected by the light condition. We apply the support vector machine to the HOG in order to detect a parson in the image.

View full abstract

Autonomous robots with visual sensors like cameras must navigate the environment without colliding with map-features which could be static or dynamic. Unlike the static features, the dynamic features needs to be detected in real time, quickly. This paper presents a parallel implementation of dynamic feature detection on embedded robot platforms. The proposed algorithm uses robot profiles which sets the frame-rate and resolution of camera based on speed of the robot. The proposed algorithm is accelerated on ARM's NEON SIMD engine and can detect dynamic features of the environment in real time.

View full abstract