In this research, we present a novel method to classify hand shape given a sequence of images capturing a hand manipulating an object obtained by a RGB-D camera. A finite set of typical hand shapes and object shapes are defined and the problem is formulated as to jointly estimate a pair of hand shape and object shape classes in each frame. Hu's invariant moment is calculated for hand region and object region separately and they are concatenated to become a single feature vector. Support Vector Machine is used to estimate the class of hand shape and object shape in each frame locally. Then, dynamic programming is applied to estimate the globally optimal object shape as well as the transition of hand shapes. Experimental results show that the proposed method outperformed the conventional methods.

抄録全体を表示

In this research, we study that information acquisition from guide sign in the indoor environment, such as airport, building, and station premises. The information helps robot to understand the meanings of environment, and to decide the path to the destination. Generally, guide sign hanged up on the ceiling consists of an arrow and some words. Our strategy is that the robot detects only the arrow, and the words around it. In the future, we will make a mobile robot to create topological map. In this paper, we try to use machine learning for the detection, assumed a guide sign place, and estimation examines indicating direction of guide sign.

抄録全体を表示

Recently, the research of using flying robot is gaining attention. The activity of flying robot is expected in various fields for example transportation, safety and security as well as search and rescue. In this paper, we propose a method to automatically detect a targeted thing by analyzing images provided from drone. In this study, the target will be human. This method can be applied to search for lost children and aged people. The characteristic to use for the detection is information of the color of the head and the color of clothes. In addition, we developed autonomous flight using GPS and also developed the system which detect the targeted person.

抄録全体を表示

We have been working on estimation of golf ball rotation between two images having only dimple patterns as clue. This paper describes a robust method of edge detection for a golf ball image using Gabor filters. Experimental results of rotation estimation show effectiveness of the proposed method.

抄録全体を表示

This paper describes an active stereo vision system on the basis of human binocular motion. A vision sensor is usually used to observe a stationary object, not a moving object. To detect a moving object, the active stereo vision system that is developed in this study is employed. The stereo vision system tracks the moving object by using a particle filter method, and measures the distance of the object by using the triangulation method. Through distance measuring experiments, the effectiveness of the system is confirmed.

抄録全体を表示

We propose an object recognition method that is based on the Multiple Instance Learning (MIL). The recognition method is able to recognize an object from the uncertain instruction of a human. Until now, many previous object recognition methods have needed many images with simple background in the training phase. However, this recognition method will be able to detect a target object from images with complex background. Through object recognition experiments, the effectiveness of the proposed object recognition method is confirmed.

抄録全体を表示

Autonomous robots that replace humans are currently in high demand in the manufacturing sector worldwide. However, these robots cannot be used in unknown environments. In this search, we aim at improving the camera calibration accuracies in order to accurately calculate the shape and position of an object from the image information obtained by two cameras.We focus on the positions of two cameras. In this reserch, we propose the optimal positions of two cameras by taking into consideration calibration accuracies. First of all, we examine the influences of the positions of two cameras for calibration accuracies. Next, we determine the optimal camera position by considering pareto optimization. After that, we verify the effectiveness of the positions of two cameras by random-bin-picking accompanied by searching task and grasping task.

抄録全体を表示

With the rapid improvement of the computation performance recently, pattern recognition has entered the generation of big data. Many different kinds of features and machine learning model have already been applied on the implementation based on big data. This paper proposed an approach of underwater creature recognition using SIFT feature and Bag-of-words model. SIFT feature, which considered to be a robust local feature, is introduced into deal with the deformation of underwater creatures. For machine learning and classification, Bag-of-words model is used in this paper to create the classifier, in a form of histogram. Evaluation experiments are executed and proved the effect of the system implemented in this paper.

抄録全体を表示

We describe a location and pose estimation of a robotic head based on three dimensional coordinate of three point markers. Computation for the location and pose estimation is carried out at 30 fps, which is a frame rate of camera, and with low latency by using FPGA. In experiments for accuracy evaluation, an error in pose estimation was less than 3 degrees. This can be used for a pose estimation of humanoid head by combining complementary other methods such as one with inertial sensors.

抄録全体を表示

An intermittent frame-by-frame tracking camera is constructed for motion-blur-free and high-brightness video shooting of fast-moving objects. To observe fast-moving objects in view, we developed a frame-by-frame tracking prototype camera, which consists of a real-time highspeed vision system and a high speed tilt piezo mirror. Our camera system demonstrate the effectiveness of our tracking camera for motion-blur-free video shooting including outdoor without decreasing the exposure time of the vision system.

抄録全体を表示

This paper presents a viewpoint planning method for an active object recognition by using the shape information of objects. The proposed method uses the certainty that a given object is classifed into one category. The certainty is represented by a probabilistic model. Moreover, the categorization possibility of unknown objects are quantied by the entropy indicating the ambiguity of the recognition. When an object is observed, the certainty is calculated by using the measured data. For each viewpoint, the entropy is computed based on the certainty. We choose as a next viewpoint the viewpoint with minimum entropy. From the experimental results using real objects, the proposed method can achieve en efficient object recognition compared with the method which selects viewpoints randomly.

抄録全体を表示

To observe multiple scenes in completely different views, we developed a galvano-mirror-based multithread active vision system that can determine its view direction frame by frame at 500 fps. On the active vision system, multiple pan-tilt-camera control processes can be multithread as 2 ms time-slicing processes; our system can metamorphose into two virtual pan-tilt cameras at 250 fps and twenty-five virtual pan-tilt cameras at 20 fps. Several experimental results demonstrate the effectiveness of virtual pan-tilt cameras control for video tracking and recording of multiple moving scenes without using multiple cameras.

抄録全体を表示

In this paper, we propose a elevator detection independent from map data. The elevator generally have a lot of designs such as the simple in the station and the decorated in the department store. It seems that the elevator has few common features, but door and button are equipped with the all ones. The former has a certain width and depth between the wall, the latter is the distinctive shape (the plate and the button). Thus, we recognize these two features using by the LIDAR data and the camera image. The algorithm achieves 90% detection without the map, which demonstrates that our method is generic and robust with variations of color, texture, lighting condition.

抄録全体を表示

It is important to observe deep sea for using sea marine resources. However, observation is difficult. Because, deep sea is an extreme environment that has high pressure, darkness and radio attenuation. Recently, Autonomous Underwater Vehicles (AUVs) are used deep seafloor exploration in order to photograph seafloor. But, the images that photographed by AUVs has low color signal by light attenuation. In this paper, we have proposed a method to enhance color in deep seafloor images. As a result, the proposed method has shown enhanced colors in each image pixel.

抄録全体を表示

Personal service robots are expected to help people in their daily life in a near future as we are facing an aging society. We deal with tasks of searching for target objects in a distant place instead of the user. But in general, a target object may not he visible clue to other objects on it. It is inefficient to pick up and check all objects in such a case. In this research, we deal with cuboids in a howl, and develop a. system that can effectively search for and pick up hidden target objects by estimating the state of occluded areas which are covered by others. We estimate the sizes and the poses of detected objects, calculate the occluded areas behind them, and determine if a target object exists in each area. We conducted experiments for several cases and confirmed the effectiveness of the proposed approach.

抄録全体を表示

Unmanned Aerial Vehicle (UAV) is prevailing to shot objects which is located in the difficult place to enter into. In this work, the high-speed active-vision system that can track target in real-time is mounted on UAV; then target tracking system from the air is developed. The system is verified through field experiment to track a ball played by a man on grassland.

抄録全体を表示

This paper describes a travelable area detecting method using a 2D laser range scanner sensor. The scan data of surrounding terrain obtained by swinging the laser range scanner sensor are plotted on a 2D image. The travelable and non-travelable areas can be detected from the image based on size of the obstacles such as curbstones, bumps and holes in the road, and so on. The experimental results show that width and height of the obstacles are possibly measured with the proposed method.

抄録全体を表示

We reported on the self-localization of autonomous outdoor mobile robots based on detecting the edge of pavement in camera images. The detecting technique of utilizing the difference in color between the edge and pavement was proposed and verified, to realize edge detection that is not affected by the sunshine condition of pavement in the captured image. The experimental results showed that the self-localization based on the edge of pavement in camera images is useful for outdoor mobile robots.

抄録全体を表示

Small humanoid robots can enter the narrow space, so there is hope that they work for lifesaving at the time of disaster or house-sitting in the home. In order to work enough in such spaces, the robot must move freely over obstacles like stairs or heaps of nibble. In this research, as the robot can move freely we propose to develop the remote operation system. Kinect that is a high-performance camera is mounted to the robot, and the operator recognizes the surrounding environment though Kinect. And operator can control the robot, the operation is reflected to the experimental robot. As the first step, we started recognition of surrounding environment and the body movement with Kinect.

抄録全体を表示

Target tracking is one of the important functions for autonomous mobile robots. In the environments where a target-tracking system is used, there may be many people and objects. When the system is applied in such environments, occlusion is a challenge for achieving target tracking. In this paper, an occlusion handling method for target tracking with a mobile robot is proposed. By using a stereo camera, both color and location information is used for target detection. During target tracking, the state of occlusion is determined. According to the state, an appropriate tracking process is applied. The effectiveness of the proposed method is verified through target-tracking experiments. In the experiments, the proposed system is compared with the target-tracking system without the occlusion handling method.

抄録全体を表示

One of the basic functions of a porter robot is to recognize and to follow its user autonomously. Though various studies has been performed, a remaining common problem is to keep recognition of the user steady. Previously, the authors have examined such a user recognition and following method using an RGB-D camera and the well-known "skeleton tracking" function. As a result, shortcomings have been clarified such as the recognition can be lost easier in a short range, which make it hard for the robot to pass around a blind corner following its user. Thus, in this report, the authors newly examine a user recognition and following method using the "user tracking" function of an RGB-D camera which is known to be simpler but faster and more robust. First, the validity of the scheme is studied through experiments where the robot is to follow its user that moves straightly. Secondly, a camera-panning mechanism is introduced to enable the robot to pass around a blind corner not losing sight of its user and avoiding collisions to walls. A prototype panning mechanism is fabricated using a 3D printer, simple algorithms are implemented to realize a panning-to-follow function, and experiments are carried out to show feasibility of the proposed scheme.

抄録全体を表示

This paper describes a method of landmark selection from image streams captured by a camera mounted on a mobile robot. To select stable visual landmarks for mobile robots, two measures to evaluate the landmark "visibility" are considered. After that, under the assumption that robots are able to seek their feasible landmark actively, repeatability of landmarks is evaluated. Weighting techniques using feature position relationships is proposed, and landmark selection criteria using variation coefficient is defined. Proof experiments using a real mobile robot shows the effectiveness of the proposed method.

抄録全体を表示

We have been developing a guiding robot to be used in the university campus, where it is definitely necessary to track a specified person among many people. In this study, we propose a tracking algorithm by the combination of simple detection by a laser range scanner and HOG features by an RGB camera to achieve a fast tracking. Some experimental results will be shown.

抄録全体を表示

A compact sized indoor mobile robot is developed, which consists of an omni-directional platform and two 6DOF manipulators. A path planning method is proposed, in which an inscribed circle of adjacent three way-points is utilized. A generic object recognition method based on its shape concept is proposed. In this article, not only size information but also aspect ratio information and other object-specific information, e.g., a cup has a hollow space, are taken into account as object shape concepts. The recognition rate was improved by applying the proposed method, i.e., approximately from 50 to 80% in case of small mag cup.

抄録全体を表示

This reserch aims to develop a 3D map generation method by a mobile robot for outdoor environments. A robot moving around in an outdoor envirionment needs to detect obstacles and recognize traversable regions. We develop a 3D map generation method using a 3D laser range finder for a wide outdoor environment. It mainly uses an ICP algorithm for estimating ego-motions. When the ICP algorithm fails, the method uses a plane matching-based ego-motion estimation or visual odometry. When we make a map in a wide area, accumulation of ego-motion estimation errors is not negligible which will distort the resultant map. We thies applied a application of pose graph SLAM to improve the accuracy of the map.

抄録全体を表示

This paper proposes a two stage target tracking algorithm for a Unmanned Ground Vehicle(UGV) to follow a small multirotor helicopter hovering in the sky. Both the UGV and the helicopter have GPS receivers to obtain their global position, but normal GPS readings are usually distorted by noise that has a several meter standard deviation, which deteriorates the tracking performance. In order to overcome this, the proposed method utilizes GPS coordinates of the UGV and the helicopter to guide the UGV roughly close to the target, and it switches the sensory information to the visual cues to estimate the UAV once the helicopter comes into the field of view. The proposed approach was verified by numerical simulations.

抄録全体を表示

By the prosperity of Internet circulation, the net mail-order firm must send out a product more quickly. To cope with this, introduction of the Automatic Guided Vehicle is pushed forward in the distribution warehouse of each net mail-order firm. It is called AGV for short. AGV is an automatic conveyance chassis. AGV is given instructions in detecting magnetic tape laid by floor and a magnetic stick, and it can run according to information from floor automatically. It plans the reduction of personnel expenses by automation and speedup of the conveyance or the automation to use this AGV for the transportation of the product in the warehouse. In this study, an image-processing module acquiring the information of the position and the information of the position gap from an image of the two-dimensional bar code acquired is developed.

抄録全体を表示

This paper describes a new camera arrangement method for bird's-eye view systems of teleoperated robots in case of camera troubles. Our proposed system provides a bird's-eye view image that helps an operator understand the surroundings easily and teleoperate the robot more accurately. The image is created by synthesizing multiple images captured by the cameras that are attached to the teleoperated robot. With a conventional method, if a camera does not work well, blind spots occur in the bird's-eye view image. In this paper, a method of camera arrangement that improves reliability of bird's-eye view systems is proposed, and the experimental results show the effectiveness of our proposed method.

抄録全体を表示

In this study, we proposed the complementary metal oxide semiconductor (CMOS) electronic circuit based on the frog vision system. The frog vision system constructed with the tectum and thalamus can recognize the simple shape such as the rectangle and square. The frog vision system is the simple structure. The simulation results with the simulation program with integrated circuit emphasis (SPICE) showed that the proposed circuit can recognized the simple shape. In the future, the realization of new vision sensor can be expected by using the proposed circuit.

抄録全体を表示

In this study, the simple sound localization circuit was constructed based on the biological auditory system. The circuit was simple structure. The proposed circuit for detecting the sound localization was evaluated by experiment. The experimental result showed that the circuit can generate the output signal for detecting the position of the sound of the target. The proposed circuit and the circuits based on the vision system were connected with the mobile robot. It was clarified from the measured results that the circuits can control the mobile robot. By applying the circuits, we can expect the realization of the novel systems such as monitoring system, collision avoidance system and robotics vision system in the future.

抄録全体を表示

This paper explains a development of an environment measuring system mounted on leg-wheel hybrid vehicle for rescue mission. Recently, there are many studies of rescue robots reported in various conferences in the world wide. The most robots aim at searching victims of some disaster. This study also aims to development of a rescue robot based on a quadruped walking robot, although the most rescue robots are crawler-type. The quadruped walking robot is equipped with rollers at the tip of the each leg. It performs roller-skates on a flat ground movement by utilizing the flexibility of the legs and walks on rough terrain as a walking. This paper also shows the three-dimensional maps as results of the 3D scanning experiments with Laser Range Finder mounted on the robot.

抄録全体を表示

The purpose of this study is to identify objects by semi-supervised learning based on clustering of 3D point cloud obtained by a laser scanner. In this study, by identifying general name of the object, reducing labor of creating learned data for the identification is contrived. Conventionally, a discriminator is manually trained by presenting each target object together with a corresponding label one by one. If the number of object to be learned becomes larger. labor for training the discriminator increases. Therefore, manual training is avoided and training method reducing the labor is proposed in this paper.

抄録全体を表示

This paper presents an online incremental method for object discovery by a mobile robot equipped with a LiDAR sensor. Obtaining object information from the real world is important for an autonomous mobile robot to enable it to understand its environment. The paper focuses on gathering object information as a point cloud from a mobile robot. In contrast to existing systems using dense 3D data, we propose an online incremental system using sparse point cloud data at each frame. Experiment results show the system effectively extracts surrounding objects.

抄録全体を表示

This paper proposes road recognition using a laser scanner. In this study, road attributes are recognized by using a laser scanner is mounted on the robot. Road attributes are recognized using road surface shape and received light intensity that can be obtained by measuring a laser scanner towards the road surface. To recognize several road attributes using the two types of measurement data, feature values are calculated. And each road attribute is modeled by Gaussian distribution using the feature values within a certain range. Recognition the road attributes are based on the Mahalanobis distance between each model and the feature vector calculated from the measurement data. Using this method, We experimented at outdoor.

抄録全体を表示

This paper reports a modeling of artificial upper extremity to analyze the compression damage caused by newly developed nursing-care equipments like care robots. In the process of risk assessment, criterion are required to evaluate the significance of injuries caused by such robots. Suitable criteria about compression damage is not available at present. Biomechanical upper extremity is created in order to propose damage criteria. Specifications of the limb are illustrated in this paper.

抄録全体を表示

It is necessary to grasp human stop operation characteristics to clarify the most suitable placement condition of the emergency stop device. The stopping performance of a robot has an influence on the placement condition of the emergency stop device. A method to measure these two stop characteristics totally is suggested by using 3D motion analysis, and particularly, a movement trace and time of human stop operation are measured and evaluated.

抄録全体を表示

The passive dynamic control (PDC) is a mechanical system control method based on an inherently safe design means using a brake mechanism effectively. Since the PDC follow-up control can be envisaged as gradually narrowing down the wide area around a control object to enclose it toward the target trajectory, the authors designate this control as "enclosing control." This paper describes the application of the enclosing control to a rotary pneumatic manipulator. First, the vision of the enclosing control is explained. Next, Experimental setup that uses two PDC pneumatic cylinders that installs electromagnetic friction brake and load cell in pneumatic cylinder is shown. Then, in a sine track trajectory follow-up control experiment and a circular trajectory follow-up control experiment using a 2-link manipulator, the effectiveness of the enclosing control is demonstrated in comparison with the PI control and the sliding mode control.

抄録全体を表示

In this paper, we proposed a stiffness changeable joint mechanism with a fail-safe function. The mechanism can change its stiffness to three states, elastic-solid-elastic, according to the safety and task situation. Each stiffness state can be switched by positioning a stiffness changing part. The design parameter was decided based on the kinematic and geometric condition of the mechanism and confirmed its validity through experiments using the prototype.

抄録全体を表示

The objectives of this study were to develop and validate a human body finite element (FE) model, which can estimate effects of muscle tones on the whole body responses with the relatively low computational costs. A whole body occupant FE model of the Total HUman Model for Safety (THUMS) ver.3 was integrated with 262 skeletal muscles. Deformable material models were assigned to all body parts of THUMS model in order to evaluate human body kinematics during impact scenarios. Each muscle was modeled using Hill-type muscle models with 800 muscle-tendon compartments of 1D truss and seatbelt elements covered whole joints in the neck, thorax, lumber, and upper and lower extremities. Muscle architectural and kinetic properties were validated in terms of muscle moment arms and the maximum isometric joint torques over a wide range of joint angles.

抄録全体を表示

This paper describes the development of a 5 DOF human-friendly robot ann. A new collision force suppression mechanism that is operated by pressure air is installed in each joint of the arm. Also, the arm is covered by an air cushion bag. When a collision between an object and the arm occurs, the collision force is passively suppressed by the collision force suppression mechanism and the air cushion. Using the ann, collision experiments are conducted, and the effectiveness of the collision force suppression mechanism is confirmed.

抄録全体を表示

This paper describes a collision force suppression mechanism that designed on the assumption that a robot collides with a human in our environment. The suppression mechanism is designed and analyzed using SolidWorks 3D CAD software. The suppression mechanism consists of a clutch gear, an air cushion bag, fixed guides, linear ways and tension springs. The mechanism will be operated even if the electric power grid is shut down and a sensor is broken. Depending on the tension spring constant, the operation condition of the collision force suppression mechanism is changed.

抄録全体を表示

Safety in human-robot coexistence systems are needed while being intensively discussed in the industrial manufacturing robots. In the future , however it will riot be practically applicable to monitor in every foreseeable situation the human-robot interaction for securing safety by use of safety-related vision-based protective equipments. Risk assessment shall be done in consideration of more close human-robot safety space management. In the study, we apply the concept of manipulability ellipsoid to that for an emergency stop to analytically the collision probability between a robot endeffector and a human eye taking avoidance actions of both human and robot actions into account.

抄録全体を表示

This paper presents a passive joint using shape memory gel. Compliance is effective for the robot arm when it touches the environment. Low stiffness robots are effective for contact tasks and the high stiffness ones are effective for precise positioning tasks. The variable stiffness robots can be used for the both tasks. The shape memory gel has high stiffness at room temperature and low stiffness at high temperature. We have developed a one-degree-of-freedom passive joint using shape memory gel. The passive joint has been designed so that it is attached to the wrist of the robot arm. A basic experiment is performed to show that compliance of the joint can be changed.

抄録全体を表示

This paper describes an evaluation study on frequency characteristics of Ionic polymer-metal composite (IPMC) sensors with Li^+ and Na^+. IPMC was composed of Nafion N117 as an ion exchange membrane and 5times gold plating electrode. To understand the time-domain frequency characteristics of the sensors, the sensor signals were evaluated by wavelet analysis method in air. As a result, the signals with Na^+ was larger than those with Li^+. It is indicated that the IPMC sensor with smaller cations has a kind of noise-reduction effect more effectively than that with the large cations because the size of cation affected the driving characteristics of the IPMC sensors.

抄録全体を表示

We propose the fabrication process of an IPMC-based soft gripper using a plasma process which is one of the most general MEMS technologies. This method enables the fabrication of the soft gripper from a sheet of IPMC. Also, this gripper can operate in solution. As a result, the operation of the gripper fabricated by the proposed process in solution was confirmed. In order to evaluate the usability of the gripper, the bending performance and the generative force under a constant voltage appliance were measured respectively. From the results, the soft gripper was able to grab a small sample, whose diameter and weigh are 1.0 mm and 12 mg, when 3.5 V was applied in DI water. These results show a feasibility of providing a miniaturized soft gripper using IPMC which has a novel structure

抄録全体を表示

ER (electro-rheological) gel is one of functional materials, which adhesive force increases when an electric field is impressed. We aim at development of force display devices which generate rotational and translational resistive forces using ER gel. The way to apply an electric field to ER gel is adopted an one side pattern electrode. One side pattern electrodes need little restriction on the environment except insulant. However, shear stress characteristics of ER gel installed to the one side pattern electrode have not been well known. In this study, we measure shear stress characteristics of ER gel on one side pattern electrode in rotational and translational movement.

抄録全体を表示

We have developed a soft PVC gel artificial muscle which has a high performance. In this study, in order to create contraction type sheets using the same characteristics of a PVC gel in electric field, we investigated the principle of the contraction type sheets, and we proposed the fundamental structure and confirmed the feasibility of the contraction type sheets.

抄録全体を表示

We have developed a structure of a contraction type actuator and investigated various characteristics of this actuator. In our previous study, we have conducted a long-term experiment to evaluate the durability of the PVC gel actuators. It was found that the displacement has a variation of decrease. In this study. we investigated the mechanism of the decrease of displacement by changing the electrodes type of the actuator. And we found there was a difference between the types of electrodes.

抄録全体を表示

In a previous study, we have proposed a new approach to evaluate the basic characteristics of PVC gel actuators, and found that the accuracy improved compared with the conventional method. In this study, we evaluated the basic static characteristics of PVC gel artificial actuators based on the new approach and modeled the static characteristics based on the Hill's muscle model as one control element for this actuator.

抄録全体を表示

SMA actuators have light-weight and big power, and especially their powers and strengths of thickened it are increased and then Self-sensing control become more difficult by their electric resistance decreased. SMA actuators have big force to weight ratio. When it is heated and the temperature reaches the transformation point, the phase changes from martensite to austenite and the length and the electric resistance decreases. We can control the displacement of SMA by managing the electric resistance. Thicker SMA has smaller electric resistance, so the control is more difficult. We try to establish accurate control of the thick SMA actuator and application it to mobile robots.

抄録全体を表示