A motion control method of lifting a heavy object up to a higher position by humanoid robots is developed. Key issue of lifting motion is how to reduce the load on humanoid arms in which lowed power actuators are implemented. The use of singular postures of arms is well-known to avoid actuator saturation of the arms. We propose a motion planner based on optimization method with making better use of actuator's characteristics. By combining two different kinds of humanoid motions such as accelerating an object upward and sliding the body into under the object, we show that the method enables to transit one singular posture of arms to another while lifting the object. We verify the effectiveness of the proposed method for reducing the load on the arms through simulation and experiment.

View full abstract

In order to exert a large force on the environment beyond the limitation of actuators, it is effective to apply impulsive force. We describe the motions that perform tasks by applying impulsive force as "impact motion." In this research, a nailing task is taken as an example of impact motion. To drive a nail skillfully, it is imperative to exert large force to a target point with precision. This paper presents evaluation of optimized motion based on virtual mass dynamics model. In order to evaluate the motion, experiments are carried out using a life-sized humanoid robot HRP-2. The position accuracy of the top head, validity of the contact dynamics model and effect of impulsive force on stability are discussed.

View full abstract

This paper describes a method of walking on rough terrain for a humanoid robot by using ZMP Criteria Map. ZMP Criteria Map is the criteria of ZMP (Zero Moment Point) trajectory modification made in advance by the walking pattern generation using the model of preview control. We classify the rough terrain into concavity and convexity, and propose a method to enable changing gait to stable walking pattern by modifying reference ZMP trajectory according to the map.

View full abstract

To realize hyper dynamic manipulations such as jumping by a smart structure, authors have proposed the utilization of dynamically-coupled driving and joint stop mechanisms before. The purpose of this work is to apply the idea to humanoids to realize hyper dynamic manipulations. This paper investigates the possibility of jumping by a small commercial humanoid with newly introduced joint stop mechanisms. The motion planning for jumping has been performed with considering the capability of actuators in the humanoid and the possibility for jumping has been examined by simulation.

View full abstract

This paper treated the Humanoid motions for climbing down a ladder based on the force sensors. The robot had seventeen DOF, and force sensors were equipped on the sole of foot for grasping a relative ladder position to the robot. Fast of all, climbing down motion was roughly divided into several postures. Each posture was interpolated to make a smoothness robot motion by developed GUI tool. The recognition of contacting the sole of foot with a ladder bar was executed by force sensors. As a result, it could be confirmed that robot climbed down a ladder successfully, just like a human.

View full abstract

This paper describes motion planning of Humanoid robot for climbing up a ladder. The robot had multi DOF, just like a human, and force sensors were equipped on the sole of foot for grasping a relative ladder position to the robot. Fast of all, climbing up motion was mainly divided into several postures. Each posture was interpolated to make a smoothness robot motion. Force sensors supported with the precise contacting motions between the sole of foot and ladder bars. Moreover, recognition system of touching the hand with a ladder bar was constructed only by foot force sensors.

View full abstract

In this paper, we propose a method to convert a motion path of a humanoid robot generated by using the PRM planner to a smooth one. By using our method, we can generate the smooth path with avoiding the collision among the links and between a link and the environment. After generating a path by using the PRM, we first make the length of the path shorter by using the random sampling technique. Then, by changing the number of nodes of the B-spline interpolation function, we show that the smooth trajectory can be obtained. The effectiveness of the proposed method is confirmed by several numerical examples.

View full abstract

This paper presents a method to generate vision and task based robot manipulator trajectories. In the framework of the research on daily assistive robots the task formulated for this work consists in picking shopping goods and placing them into a cart. For object recognition SIFT keypoint tracking in combination with stereo vision is used. A collision free trajectory between two waypoints is generated based on a 3D-gridmap of the environment in front of the robot. A trajectory generation loop is implemented using resolved motion rate control and attractor coordinate frames that define the waypoints of the trajectory. The proposed method was tested on the humanoid robot HRP2-JSK.

View full abstract

Humanoid robots in daily supporting at home need to have not only behavior control system based on human instruction but also autonomous behavior selection system. To realize it, situation recognition capability is necessary. In this paper, we propose situation recognition system based on history of recognition of daily life sound and visual recognition of human existence. In this system, a key point for situation recognition is to use time series of recognition history. Also daily life sound represents environmental changes. So, humanoids can recognize robustly environmental situation through integrating comparatively long-term information obtained from real world. As an example of situation recognition, we show a case targeting a human in a kitchen.

View full abstract

Task planner provides complex motion sequence from a high-level command. In order to apply task planner to a real humanoid robot, keeping consistency between a real world status and a planner status is one of the key issues. This paper presents a method to generate planner status by using visual recognition system that utilize environment information in the task planner.

View full abstract

In this paper, we discuss wheelchair support by a life-sized humanoid robot. In wheelchair support, integration between sensory information and behavior is important to adapt the humanoid robot to the physics of the real world, environment and the user. This paper aims at the support that the humanoid robot pushes the wheelchair and the user can transfer to anywhere only by instructing where to go.

View full abstract

The rational formula and structure formula of robots were proposed. The axis direction of the motor which composes a robot is classified by Roll axis, Pitch axis and Yaw axis. When there is a motor on the body side, the initial of axis is written with a capital letter, and written with a small letter in case of opposition. The whole structure of the robot is expressed as a rational formula of the robot to show molecule structure in chemistry. SANSYRO n is expressed as a rational formula of the robot with "Humamoid-16DOF=Head-Y-(Arm=Pr-0-0)_2-Y-(Leg=RP-p-PR)_2". Human structure is analyzed, and it is described about the way of developing humanoid.

View full abstract

This paper presents development of an automatic wheelchair control system via electroencephalogram (EEC) and its experimental results in an office building environment. The developed control system mainly consists of two parts. One is an autonomous control part that achieves automatic wheelchair control based on sensor information. The other is a decision making part that realizes alternative turn direction decision only by EEG signal. The former part is applied to move hallways from a corner to the next corner. At each corner on the floor, the latter part is used to make a turn direction decision. Our experimental results show the possibility of realizing simple tasks in wheelchair control as a brain machine interface application.

View full abstract

In this paper, we build a whole-body neuromusculoskeletal network model including somatic reflex, and identify its parameters through non-invasive measurements and statistical analysis. Our neuromuscular model consists of two parts. The first part models the neuromuscular network to represent the relationships between spinal nerve signals and muscle activities, which are then converted to muscle tensions using a physiological muscle dynamics model. The second part includes the feedback loops from muscle spindles and Golgi tendon organs to spinal nerves that represent the somatic reflex using muscle length, velocity, and tension information. We demonstrate the consistency of the model by showing that a forward dynamics simulation of somatic reflex yields a motion similar to the actual response of human.

View full abstract

This paper proposes a computational model for phoneme acquisition by infants. Infants perceive speech not as discrete phoneme sequences but as continuous acoustic signals. One of critical problems in phoneme acquisition is the design for segmenting these continuous speech. The key idea to solve this problem is that articulatory mechanisms such as the vocal tract help human beings to perceive sound units corresponding to phonemes. To segment acoustic signal with articulatory movement, our system was implemented by using a physical vocal tract model, called the Maeda model, and applying a segmenting method using Recurrent Neural Network with Parametric Bias (RNNPB). This method determines segmentation boundaries in a sequence using the prediction error of the RNNPB model, and the PB values obtained by the method can be encoded as kind of phonemes. Experimental results demonstrated that our system could self-organize the same phonemes in different continuous sounds. This suggests that our model reflects the process of phoneme acquisition.

View full abstract

CPG(Central Pattern Generator) exist in the brain of animal. It makes rhythm for motion of animal. CPG have a feature that if we connect many neural oscillator each other, rhythm that outputted from one. It is interesting for distributed control to apply neural oscillators. In this paper, we report about development of distributed control system by neural oscillators. We build neural oscillator network, which consists of the Matsuoka model, is introduced to realize the locomotion of a multi link mobile robot. We developed distributed control board for DC motor control by PIC18F452. The robot can move forwards and change direction by adjusting command parameters from upper layer.

View full abstract

Our goal is to create an autonomous robot which can communicate with people for a long time. In this paper, a emotional motion generation algorithm constructed of an endocrine system model and a neural network algorithm is described. We carried out an experiment to survey the adaptability and flexibility of the robot's behaviors. As the result, the robot with our proposing algorithm could adapt to the simulated environment including a predetor which discharges the robot's battery.

View full abstract

We humans learn behavior skills through repetitions of own behavior experiences. In the brain science perspective, we speculate such skilled behaviors might be acquired in inferior parietal lobe through the repeated sensory-motor experiences and propose a brain model. We utilized two neural networks. One is a Continuous Time Recurrent Nenral Network (CTRNN) which learns the trajectory of joint angles to achieve the task. Another is Recurrent Neural Network (RNN) which learns the relation of muscle activity to joint angles. To test the proposed model, we applied it to the task of reaching for an object using real robot with a tendon-based actuation mechanism.

View full abstract

Kohonen's self Organizing Map (SOM) is one of the candidates for construction of somatosensory area which have a topological map corresponding own body. Information representation in SOM is continuous value as rate coding although neurons deliver information by spiking. Recently, Spike timing dependent plasticity (STOP) learning rule is proposed from researches of brain physiology. In this paper, for preliminary experiment, we have examined whether a somatosensory map for simple musculoskeletal system is constructed by spindle output through STOP. In the result, we confirmed that selective neurons for each input emerged, where topological map is not constructed.

View full abstract

We have accelerated a robust model-based 3D tracking system by programmable graphics hardware to run online at frame-rate during operation of a humanoid robot and to efficiently auto-initialize. The tracker recovers the full 6 degree-of-freedom pose of viewable objects relative to the robot. Leveraging the computational resources of the CPU for perception has enabled us to increase our tracker's robustness to the significant camera displacement and camera shake typically encountered during humanoid navigation. We have combined our approach with a footstep planner and a controller capable of adaptively adjusting the height of swing leg trajectories. The resulting integrated perception-planning-action system has allowed an HRP-2 humanoid robot to successfully and rapidly localize, approach and climb stairs, as well as to avoid obstacles during walking.

View full abstract

In many computer vision applications such as surveillance, targets to detect and measure move. We have proposed a stereo camera system that focuses on moving regions. In the system, what is called the "correspondence problem" of the stereo cameras is solved by restricting the stereo matching process to moving regions. In this paper, the fundamental algorithm to realize robust measurement of stereo images by detecting moving regions with each camera and then applying stereo matching for the detected moving regions is verified, and detection of features of moving regions using the proposed stereo system is discussed and evaluated by experiments.

View full abstract

In this paper, we present a method to obtain the depth of the building using a satellite image of Google Earth and GPS receiver. Maps with 3D information are useful for robot navigation. This method is based on "3D Street View", which is a 3D model of the building automatically created with laser range finder and fish eye lens camera.

View full abstract

A method of range image sensing by integration of active and passive methods using multiple spotlights is proposed. The active stereo method using multiple spotlights can sense fast and robustly, but its measurement is sparse. The passive stereo method using stereo camera realizes dense measurement, but it cannot sense fast. Using the information of active stereo sensing makes the measurement time of passive stereo sensing by a stereo camera shorter and makes its accuracy higher. This paper explains the improvement of the range image sensor that integrates active and passive methods using multiple spotlights and a stereo camera.

View full abstract

This paper describes 3D environment modeling and camera pose estimation system by using a single camera. Because a wide angle camera provides a wide field of view at once, it enables the system to acquire reconstruction results stably. Moreover, a camera which has advantages of small size and light weight, is easily considered to apply the system to a portable sensor. We constructed environment modeling and camera pose estimation system by means of SFM approach (8 point algorithm, bundle adjustment and motion stereo). Graph slam is also adapted to smooth the reconstruction results. Experiments were performed in indoor environment, we comfirmed the capability of the reconstruction ability of a wide angle camera.

View full abstract

As for the pedestrian detection method, It ever proposed various one. But it can't detects the pedestrian in dynamic environment. So we researched one on mobile robot that is dynamic environment. In this paper, we generate 3D optical flow from gotten range image and intensity from stereo camera. It picks out double support phase and single support phase by using variance of this flow vector. It detects the pedestrian by using row of double support phase and single support phase. It is predictably effective because a variance of this flow vector is invariable on mobile robot. It could detect the pedestrian with the accuracy of 79.4% on wheel-type mobile robot.

View full abstract

The goal of this study is to make pedestrian supporting geographic information systems by low cost. For the purpose, we mounted stereo vision camera on electric wheelchair that max speed is six kilometers per hour and obtain stereo vision images. We calculate candidates of borderline of sidewalk from each image by off-line, and recognize the type of the borderline (for example, curb, gutter, wall, white line, and so on). From these results from multiple images we calculate width, traffic, and road surface condition of each sidewalk. We applied this algorithm to the images obtained from low-speed outdoor vehicle (based on commercial electric scooter) and show the efficiency of it.

View full abstract

This paper deals with sensing using laser range finders and improving or altering their characteristics using mirrors. The idea that is presented here is to use one or multiple mirrors to change the direction of the laser ray, thereby resulting in some characteristics that are not available to a simple laser range finder, such as: scanning in multiple planes, which can be useful in e.g. 3D scanning; or multiple observations of the same area in one scan, which in turn gives an increase in the reliability and measurement resolution. The methods are explained and sample applications are presented, including the use in the laser range finder based touch screen.

View full abstract

This paper describes a scanning method for taking three dimensional reflecting points in real space. We made up a three dimensional scanner that a body of two dimensional SOKUIKI sensor is swung by roundly swing mechanism. We took actually reflecting points in real space by using the three dimensional scanner. The reflecting points corresponded with size of the real space token by the three dimensional scanner.

View full abstract

This paper described an all-sky 3D scanner. This sensor can scan surrounding 3D shape and color, because a 2D range sensor and multiple CCD cameras are rotated in the direction of a pan angle by a turntable. The information by obtained the all-sky 3D scanner is the point data and the color. By modeling, the point data and the color data were united and the 3D surface model was manufactured.

View full abstract

This paper describes a new three-dimensional environment recognition method for 3-D mobile robot that can walk up and down the stairs. The prototype system named MEROS (Mobile Environment Recognition Sensor) was developed. The system consists of two-dimensional scanning laser range finder (URG), Z-axis revolving mechanism and a computer. The URG measures the distance of every angle interval in a scan plane. The revolving mechanism makes the sensor rotate and it is positioned in the uniformly space. The computer processes the sensor direction, the scan angle and the measurement distance. The procedures are as follows. 1) The extraction of the feature quantity of the vertical scanning plane. 2) The extraction of the horizontal feature quantity concerning multiple scanning planes. 3) The recognition of the environment according to the consolidation of the information. When the system finds stairs, it can measure the relative location between the sensor and the stairs and the geometries of the stairs.

View full abstract

In this paper, we present a system to automatically create a 3D map of city's buildings. The system consists of a laser range finder to measure distance between the sensor set and the building; and also a fish-eye-lens camera to measure the height of the buildings while capturing the texture of the model. Our system allows not only capturing the geometry of the buildings, but also how the buildings would look like on a street-level view. This allows for the creation of a 3D model that is suitable for navigation purposes.

View full abstract

This paper reports a sensing point planning for mobile robots to obtain 3-dimensional environment information which includes range and visual data. Generally, to avoid an occlusion problem and a low resolution problem, it is necessary for robots to sense in different places. Therefore, a suitable sensing point planning is required to reduce redundant sensing. This problem is called "Art Gallery and Illumination Problems" in 3-dimensional environment, and our approach is to solve the problem with consideration of both valid-sensing-area and partially-unknown environment. In this paper, we explain our planning-approach and introduce some initial experimental results.

View full abstract

A digital pen using three dimensional instrumentation technology by the inertial navigation method is developed. This paper describes the estimation method for sensor offset aiming at accurate pen tip trajectory and the verification of the effect.

View full abstract

This paper presents the indoor object management system implemented in sensor-embedded environment using active Radio Frequency Identification (RFID) technology. The proposed system provides users with functions to management targeted objects in the room, monitor their locations in realtime, view and search for the movement history of objects from log database. The proposed object localization method is based on the analysis of RF signal strength retrieved from RFID system and position tracking of user in the room using floor-distributed pressure sensors. The experiment results have shown that the integration of sensors can significantly reduce the effects from external interferences that affect the performance of RF signal strength analysis, and result in the improvement of indoor object localization accuracy.

View full abstract

This paper describes a method of fault diagnosis of in-vehicle sensors for wheeled mobile robot. We handle scale fault of three internal sensors (two wheel-resolvers and one gyro). The scale fault is diagnosed based on estimating the robot velocity with an external sensor (laser range sensor) via a point-to-point scan matching method. We also detect the fault of the laser range sensor based on the laser scan matching; if the error related to the laser scan matching becomes a large value, the laser range sensor is decided to be faulty. The experimental result validates the effectiveness of our method.

View full abstract

In this research, we propose the robot technology (RT) ontology for Kukanchi (Interactive Human-Space Design and Intelligence) to make robot understand the user's intention, and provide the appropriate service according to the situation. Proposing technique of RT ontology can be used for the infrastructure technology of the space information structuring. In this paper, we describe the outline of RT ontology and topics about this research.

View full abstract

In general, verval orders from a user are simplified or shortened and contain ambiguities. For example, an order "where?" issued by a user may mean the place of carrot when he/she is cooking or may mean the place of sugar when he/she is making coffee. In this paper, we propose a context-aware work support system, which provides appropriate services depending on human action context. The ambiguities of user orders are resolved by recognizing human actions using sensors embedded in the environments. We conducted experiments with subjects and confirmed that the proposed system provides appropriate services according to user action context.

View full abstract

There are various approaches in robot-human interaction. In one approach, a robot is used as media. We will expect the robot to give us information and behave like as a presenter or a teacher before long. In order to indicate information properly in this case, the robot needs to know who is interested in what. In other words, measurement of interest is important. In this paper, we propose a map showing the distribution of people who interested in a certain object to help the robot's decision. We conducted a experiment in the real environment and evaluated the resultant map.

View full abstract

Gesture motion has different means according to circumstances. Human recognize other person's intention or attending points from gesture, face direction, situation and so on. In this research, a consistent interface to give the necessary environment phase division and necessary information is constructed according to the scale on which the environment is called the house, the store, and a shopping street. The user's understanding easily from the use of three dimension visualization space as Virtual-City to the interface in the sight, and using it easily become possible. In addition, the installation of the marker etc. achieves the instruction by the gesture in natural shape without the necessity for the interface by using the vision board.

View full abstract

In this research, we propose a method to estimate human hand position. We aim to develop simple 3D models with Spatial Memory and in order to realize such system in the Intelligent Space, human hand localization is needed. Now we are using the ultrasound localization system for the purpose but it sometimes can't get actual position. When transceiver inclines or blocked by an object, the localization system can't give us the position. In this paper, we will describe the human hand localization system by using inertial sensors and also the way of evaluation.

View full abstract

In this paper, Intelligent Space based human-object relations observation system is proposed and constructed. Intelligent Space is an environment that has distributed sensors for observing and actuators for acting in the space, in order to provide various services to human. There are a lot of objects in Intelligent Space and in order to offer appropriate services using these objects, not only the physical information about the object but also the object's information that occurred by interaction with human is needed. We try to define human-object relations based on the use history of objects and discuss the method of observing it.

View full abstract

It reports on the appearance of the demonstration concerning the home service with multiple robots of spreading table, conversation robot, and media mirror in near future house though it was open to the public at the end of last year in Waseda University WABOT-HOUSE laboratory.

View full abstract

Automatic detection and description of events, particularly human behavior, is one of the most challenging issues since the event interpretation is highly dependent on the target of attention which is not specified uniquely. In order to tackle this problem, we propose the concept of "cognitive ontology" as a framework of automatic decision of attention focus and event description system. The cognitive ontology is structured with conceptual units which are entity and relation, and those units enable robot endogenous attention fixation and jump based on networked Cognitive Ontology. Additionally, we introduce exogenous attention based on the difference from a predicted pattern of the observed targets. In this process, the corresponding target of attention is updated and set to event description buffers which is consist of two entities and one relation. In this paper, we developed and experimented this holistic event interpretation process according to endogenous attention, exogenous attention and determined event description.

View full abstract

I propose operation user interface in the intelligent space for the life support robot. It aims to demand service certainly easily by cooperating with intelligent space. In this paper, we describe the necessary condition, system configuration and develop a prototype using acceleration and azimuth sensor.

View full abstract

We are conducting the research project named "Robot Town Project". The aim of this research is to develop a distributed sensor system and its management system so that robots can work with humans in an ordinary environment for daily human life. However, it is difficult for distributed sensors to cover all of the environment. This paper describes a method of cooperative tracking of pedestrians by an on-board Laser Range Finder (LRF) of mobile robot and a fixed sensor. The method successfully expands the area of sensing, and reduces the occlusion. The fixed vision measures the position of the mobile robot in the world coordinate system, and the on-board LRF measures the relative position with the mobile robot of pedestrians using Kalman filter. Integrating these information, this system can track pedestrian in the world coordinate system.

View full abstract

In the research field of "KUKANCHI: Interactive Human-Space Design and Intelligence", it is difficult for the robot to acquire the data from environment corresponding to components which has various interfaces, such as Ethernet, RS-232C and USB. In this research, we construct the infrastructure to acquire the data from components using ORiN. As a result, the robots can acquire the data via common interface without supporting various hardware/software interfaces and we can re-use components in the past.

View full abstract

Robot action framework using aspect oriented concept is proposed to cope with cross-cutting concerns in robot tasks. The knowledge for concerned objects is built into the objects themselves, but the knowledge for cross-cutting concerns is written in "aspects." For example, exceptions, such as obstacle avoidance, are one of such cross-cutting knowledge. This framework makes robot software easy to be maintained and reused.

View full abstract

Ubiquitous robots consist of distributed and networked robot elements such as sensor, computer, actuator, and etc. Distributed actuator is one of the ubiquitous robot elements, which provides physical services. We are developing an actuation module which is distributed easy form. Here, communication channel between actuators and controller need to be improved in more easy dispersion of actuators. Wireless communication can distribute actuator to space easy. When huge number of actuators is distributed, necessary communication will be obstructed by communication traffic. And operating many actuators is over capacity for controller. This paper solves communication problem using by multicasting communication and the time sync system.

View full abstract

We are proposing a data structure used for describing environmental information to make various robots execute door closing tasks based on 'Kukanchi'. In this paper, we report experimental results of two different kind of door closing task with a mobile manipulator according to a uniform data structure.

View full abstract

Our group has proposed a new communication technology named "Two-Dimensional Communication" for high-density sensor networks. In this paper, we propose the method to estimate the positions and orientations of sensor nodes on the Two-Dimensional Communication sheet (2DC sheet). First, we introduce the position encoding scheme on the 2DC sheet. Next, we analyze the electromagnetic field modulation by the position code on the 2DC sheet, and propose the method to detect the position code on the 2DC sheet using the electric field sensing. Finally, we show the results of the fundamental experiment to confirm that the proposed method is useful for the position estimation on the 2DC sheet.

View full abstract