The purpose of this study is to avoid falling using changing step for humanoid robots when humanoid robots cause a large disturbance. To realize the fall-avoidance, humanoid robots detect a falling possibility. This study adopts the phase plane trajectory to detect falling and experimentally discussed about its problems. The changing step motion is realized the inverted pendulum model with two support points.

View full abstract

Humanoid robot which can work in living space is required. Because daily chores or physical cares need various contacts, a whole body contact motion of humanoid is necessary. In this research, we develop a humanoid robot with whole body distributed tactic sensor. A tactile sensor for whole body motion should be easy to maintain because it is applied by an impact and shear force. So we make soft removable tactile skins for all links separately. To realize this, we use "cut-and-paste of tactile sensors". Finally we realize some whole body contact motions of an adult size humanoid robot.

View full abstract

In many researches for a humanoid balance control, the floor reaction force operation and the center of gravity acceleration operation according to it have been discussed as a center problem. However, in order to act in a real complicated environment for humanoid, it is essentially indispensable to switch the balance maintenance actions according to time constraint like taking a step or falling down defensively. In this research, the balance return action generation method corresponding to time constraint is described. The effectiveness is shown by the balance simulation.

View full abstract

Humanoids should operate various objects in human life environments, and act skillfully contacting with the environment to support human life. It is thought that humanoid's hands that have mechanism automatically transforming into appropriate shape, and have sense of force are effective for human life support. In this paper, We describe autograsping mechanism, passive joint, and tactile sensor shell of humanoid's hand. And we show integration of a humanoid hand with those mecanism and report experiments executing some actions using that hand.

View full abstract

Musculo-skeletal humanoid has many DOFs and flexible structure, so it can adapt to environment using hardware. To realize adaptation to environment, it is important for musculo-skeletal humanoid to modify its actions by using its own sensors because it is difficult to model for its complex structure. In this paper, by vision-based reinforcement learning, musculo-skeletal humanoid Kotaro aquires crank rotation behavior with flexible spine.

View full abstract

In "Real World Information Systems Project" in "21st Century COE Program", we demonstrated research progress, called "A Living Room Of The Future". This shows an example of situations of humanoids daily life supporting, where HRP-2W, with HRP-2JSK, follows an order from a guest, serve tea, and take away the cup. We use "Reactive Server System", activating behaviors by events thrown by each module, for realization of daily task using functional abilities of grasping a tea cup, moving around the stage, visual recognition of cup location, speech recognition, sound source localization, voice synthesis and so on. In this paper, we will describe summary of implementaion, artifice and problems in deed, and will describe solutions and future prospects.

View full abstract

The improvement of the performance of the leg was done by assembling the elasticity body that muscle was modeled in the leg modeled with servomotor. Three muscles (gluteus maximus, rectus femoris and gastrocnemius) were modeled. The first stage prototype was used in the process of examining the modeled muscles. The first stage prototype is the robot composed of three main parts (Servo-motors, C form parts and box parts) and it can be made easily. After that, the effect of the muscular modeling was confirmed by using original robot SANSYRO II of our laboratory. When elasticity bodies were put, SANSYRO II could lift the weight of maximum 3.75kg (3 times of the tare mass).

View full abstract

This paper describes a LCP-based method for calculating contact forces in dynamics simulation of humanoid robots. The method is based on the algorithm proposed by Kokkevis, and we propose additional procedures that make the algorithm more practical in the reality and numerical stability. The simulation system is verified by experiments using humanoid robot HRP-2, and the result shows the efficiency and validity of the method.

View full abstract

This paper studies an optimal planning of falling motions of a human-sized humanoid robot to reduce the damage of the robot. We developed a human-sized robot HRP-2FX which has a simplified humanoid robot shape with seven d.o.f. and can emulate motions in the sagittal plane of a humanoid robot. An optimal control is applied to generate the falling motion of HRP-2FX to minimize a performance index, and the optimality has been verified by the experiments on HRP-2FX.

View full abstract

Pivoting manipulation has such advantages as dexterity and safety over other methods to move bulky or heavy objects. In this paper we aim to show that an polyhedral object can be displaced to arbitrary position and orientation on a plane. For this purpose we analyze "small-time controllability (STC)" of pivoting operations and establish a motion planning algorithm. The manipulation motion is first planned as a path of a vehicile whose steering method has nonholonomic constraint and is then transformed into a pivoting operation sequence.

View full abstract

This paper explains the construction of a behavior level operation system which enables an operator to operate humanoid robots by using natural language instructions. We report on a system using a speech recognition module built on an embedded system and the implementation of the system using RT middleware on HRP-2.

View full abstract

A motion control method of lifting a heavy object up to a higher position with 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. By combining two different kinds of humanoid motions such as accelerating an object upward and sliding the body into under the object, we propose a method that enables to transit one singular posture of arms to another while lifting the object. Simulation results show the effectiveness of the proposed method for reducing the load on the arms.

View full abstract

In order to effectively generate an impact motion of a humanoid robot, a motion generation support system is developed. The system visualizes not only a designed motion but also experimented motion. Force and torque generated in the experiment are superimposed on the experimented motion. The visualized ZMP, GCoM, force, moment and difference between the designed motion and the experimental result will help motion designer to improve the designed impact motion. A nailing task is taken as an example of an impact motion. A motion of a humanoid robot to drive a nail into a wooden plate with a hammer is designed using this support system. The details of the motion generation support system and the nailing experiment performed by a humanoid robot are presented.

View full abstract

Superior navigational ability in everyday environments such as homes and offices is one of the primary incentives behind humanoid robotics research. Yet their ability to step over or upon obstacles remains unexploited if algorithms designed for wheeled robot navigation are also employed for humanoids. Recently, globalized strategies have been designed that do take into account this feature. Yet, due to high computational complexity, most of them remain highly time consumptive. A new approach is presented herewith that employs a human-like approach to terrain traversal. Results of its simulation on a model of our humanoid robot 'Saika-3' are also presented. The strategy is reactive, yet global and reduces computational complexity by limiting graph expansion to only useful paths.

View full abstract

Performance of human size humanoid robots is strictly limited by performance of motors. The progress of the motors has not been remarkable compared with the progress of electronics. Therefore, the great progress of the performance of the humanoid robots can not be expected, at least in the present circumstances. In this paper, a gravity compensation mechanism is designed which is applicable to general biped robots. The mechanism is expected to reduce the joint torque of the legs required to support the gravitational force of the whole body. A humanoid robot Saika-4 is equipped with the gravity compensation mechanism in the legs. To evaluate effectiveness of the gravity compensation mechanism, preliminary experiments are performed using the humanoid robot Saika-4.

View full abstract

Motion imitation is one of methods to acquire new motions from interaction with environment. Our aim of this study is to realize the body motion imitation of humanoid robot that uses only visual information and that need not knowledge such as joints or links structures of demonstrator such as other robot or a human.

View full abstract

This paper deals with the problem of determining the minimum number of degrees of freedom (DOF) a humanoid robot should have to be able to produce natural motions. A human model with a high number of DOF is used to reproduce captured natural human motions. The model's DOF is reduced gradually until reaching the boundary of the naturalness of a generated motion. An experimental study, using self-reporting approach for the evaluation of the naturalness of the generated motions, is presented and discussed.

View full abstract

Humanoid robots must maintain balance to ensure safe operation. The purpose of this paper is to introduce a balance control method for a humanoid robot based on the Reaction Null-Space Method, when a sudden disturbance is applied to the robot. Experiments with a HOAP-2 humanoid robot confirm the validity of the proposed control approach for mimicking the so-called ankle and hip strategy used by humans.

View full abstract

This paper studies locomotion planning of humanoid robots to pass under obstacles. In accordance with analyses of experimental motions of human passing under obstacle, we propose a new category of humanoid's motion that a humanoid turns sideways to pass under obstacle. Possible initial and goal configurations of the motion are calculated with humanoid robot HRP-2 model.

View full abstract

A motion planning method of humanoid robots to pass under obstacles is proposed. The motion planner consists of Motion Path Planner, which can generate statically stable motion under considering joint movable range and collision avoidance, and Motion Timing Planner, which can generate dynamically stable motion under considering ZMP conditions and maximum joint velocities. The effectiveness of the proposed method is experimentally confirmed with humanoid robot HRP-2.

View full abstract

The author has been engaged in studies of biped walking robots and bipedal humanoid robots for 15 years. The author was a technical adviser of the Sony Dream Robot Project from 1997 to 1999, and a technical consultant of the project from 1997 to 2004. The project developed bipedal humanoid robot SDRs and QRIOs. The author is a joint inventor of SDRs and QRIOs. In 2004, the author became an intellectual property right common holder of QRIO Motion Dynamics Technologies. In this paper, the author introduces motion stabilization patent technologies of a legged walking robot made right by Sony Corporation and Jinichi Yamaguchi.

View full abstract

In our previous work, a walking pattern generator based on the spherical inverted pendulum model has been proposed. It enables to generate up-and-down waist motions and utilizes singularity neighborhood in the middle of each support phase. In this paper, we propose a new method to generate a walking pattern based on Spherical Inverted Pendulum with Underfloor Pivot, SIPUP, whose pivot is set under floor. As a result, ZMP has mobility during each single-leg support phase. Experimental results show the effectiveness of our approach.

View full abstract

Real-time gait change is necessary for humanoid robots for emergency. We propose a new gait change method using modification criteria map. The stable gait change is generated by adjusting the amount of the ZMP modification according to the timing of stop command. The modified ZMP trajectory is given so that the humanoid robot can change the current motion without falling down. The modification criteria are defined from the relation between the predicted ZMP trajectory using a preview controller and the support polygon. The preview controller employs Table-Cart model and it derives Center of Mass(CoM) trajectory from ZMP reference in real-time. We make the map of relation among the ZMP modification length, the modification timing and the timing of the stop command for stable gate modification. The robot can execute the best motion referring to the predefined map. In this method, the humanoid robot can stop or change step length immediately within one step. The validity of the proposed method is confirmed by experiment using a humanoid robot HRP-2.

View full abstract

One of the issues of a biped robot is to realize jumping and running as well as walking. The compliance of the ankle joint is supposed to be a key for realizing stable running. This papar reports the relationship between the ankle joint compliance and jumping motion. An experimental result shows that the ankle joint compliance changed success rate and direction of jumping. This result suggests that the feedback controller can be constructed.

View full abstract

In this paper, we propose a step-out motion planning to stop robot movements from arbitrary positoin and velocity of the center of mass (COG). This method calculates the desired foot landing position to gain dynamically feasible recovery force, and generates the COG trajectory of step-out motion based on the optimal regulator for continuous inverted pendulum system.

View full abstract

This paper shows the details of Zero Moment Manifold control method for an under-actuated gymnastics robot. In this method, though we use a gravity compensation term at the desired posture (Off-line method), we have not aimed at a reduction of computational complexity of real-time gravity compensation (On-line method). Off-line method is used to linearize the dynamics of a gymnastics robot. Numerical simulation shows the difference of Off-line method and On-line method.

View full abstract

If multiple manipulators are placed in each others' working area, it's need to avoid collision of manipulators. To detect the collision, we propose the detection method that manipulators are approximated by some spheres. The detection check is calculated in real-time and efficiency because it's easy to calculate the distance between the objects.

View full abstract

A sensor-based navigation algorithm for mobile robot such as Tangent Bug algorithm works only for stationary obstacles. In this paper we proposed the navigation algorithm which works for moving obstacles and stationary ones with unknown environment, using a new idea that distinguishes moving obstacles from stationary ones with a distance information of the sensor of mobile robot, while moving along the boundary of the obstacles. The idea is based on the definition of an angle toward the wall surface that is called the direction vector toward the wall surface. The angle of the direction vector of each step is accumulated while following the boundary. When the total accumulated angle is over 2π radian, the following obstacle is recognized a moving obstacle. In this paper the effectiveness of this algorithm for moving obstacles is shown by some simulations.

View full abstract

This article investigates how grasping position errors affect the internal force applied to the manipulated object in dual-arm impedance control. First, how to regulate the internal force using a dual-arm impedance control scheme is described. Second, internal force errors caused by grasping position errors are analytically derived. Finally, a remedy for the internal force errors is discussed based on the analytical results.

View full abstract

In this paper, we propose a method to estimate the relative positions of multiple robots in a transportation system without using force/torque sensors. The estimation method is based on robots measured velocities using Particle Filter. The estimation algorithm is implemented to three omni-directional mobile robots, and the experimental results illustrate the validity of this method.

View full abstract

We report the development of a system that allows parents to remotely participate in the day-care activities of their pre-school children and share notes with the childrens' nurse. In this system, "Action Switch platform (AcSP)" links the PaPeRo (Partner-type Personal Robot) to the cellular phone of the parent. PaPeRo is driven by the content of the parent's e-mails. A Blog-based system that uses live pictures of the children (these data are captured by PaPeRo'camera and stored in AcSP) is used to realize intelligence sharing. We implemented this system in two pre-schools as a trial; 12 families participated. Data was collected by interviewing the families and the nurses. We report their comments, recommendations, and our conclusions.

View full abstract

This research addresses the problem of the allocation of network robots to serve multiple users in multiple areas. Service assignment architecture for network robots is described, which has the following features: (1) All statuses of a variety of robots and users are systematically abstracted in the form of 4W (Who, When, Where, What), especially the "where" is composed of the landmark and numerical or textual information. (2) From some representatives of a scenario defining a required service, an appropriate service and scenario related to the service area is determined by 4W1H matching and then executed. (3) The service history of users recorded in a database and service contents are downloaded to robots in CroSSML format. The demonstration given in NTT Core Technology Symposium and R&D Forum 2007 confirmed that services performed in different area could be provided to multiple users simultaneously by visible robots and virtual robots according to the status of users.

View full abstract

What is the key feature of a humanoid which communicates daily with people? Previous work showed that a communication robot in an office can acquire familiarity with daily chats. However, the robot could not give familiarity to all the people in the office. Then we have come to conclusion that the robot needs to adapt its behavior appropriately to current situation, have a suitable role for the environment, and give friendly impression. In this paper, we investigate how people change their impressions of the robot if it changes its behaviors depending on the situation or has a role (or a job) in an office.

View full abstract

While a robot and a human are communicating with each other, much information about the human such as gender and personality is observable by the robot through tactile information. This paper proposes a method for estimating a human's gender, personality traits and emotional state by using tactile information measured over the entire body of a robot. We conducted communication experiments between a robot and a human and recorded tactile information as well as the human's gender, personality traits and emotional state. Our results show that each of these characteristics is correlated with observable patterns in the tactile data. By building on these findings, robots will be able to estimate information about humans based on tactile information.

View full abstract

We developed a small humanoid for studies on full-body behaviors, Syokabe. The 26-dof robot has a pitching axis in each foot to rotate the tiptoe and a yawing axis at the waist and each of the two thighs. These five rotation axes make it possible for the robot to stably walk with long strides. Furthermore, the robot is suitable for frequent small turns and thus playing sports, dancing, human robot interaction and so on. We will present the mechanism and some full-body motions of Syokabe.

View full abstract

We realized a humanoid robot which kicks balls in various directions using the yawing rotation of its waist and two thighs. It is capable of three kinds of kicks, toe, inside, and outside. For example, the toe-kick can hit a ball in front of the robot towards various targets. We conducted experiments to verify that the robot can kick balls placed in different positions effectively in various directions. We tested each type of the kicks changing the yawing angle of the kicking leg and investigated the relation between the angle and trajectories. We will present the kick methods and results of the experiments.

View full abstract

Cooperative manipulation by multiple industrial robots with parallel grippers is achieved using slippage between grippers and the object. Some experiments are done to verify that all states of slippages can be realized by controlling the manipulation forces and the grasping forces applied to the object.

View full abstract

The paper describes a behavior of collective robots by applying the intermolecular potential model. The Lennard-Jones potential that is one of the models of the intermolecular potential shows a dynamic interaction of intermolecular. In the Lennard-Jones potential, repulsion and gravitation corresponding to the intermolecular distance are caused by the energy level of the molecule. The behavior generation of the robot group is tried by applying this model as the interaction model among robots. Virtual interaction is caused according to the distance between robots when this model is applied to the collective robots, and it is thought that state of the robot group can be controlled. Then, control of the collective movement of the robots is tried by using the Lennard-Jones potential through simulation.

View full abstract

In this paper, we propose a decentralized control system of multiple mobile manipulators handling a single object in coordination without using force/torque sensors. In this system, the grasping point of each robot is controlled as if it has an impedance dynamics without using a force/torque sensor, by using the identified parameters of each robot. These robots handle a single object in coordination using a leader-follower type control algorithm. We also consider the effect of parameter identification errors and propose a method to reduce the effect of them. The proposed control system is experimentally applied to two mobile manipulators, and the experimental results illustrate the validity of the proposed control system.

View full abstract

We are proposing the cluster robot system as one example of the autonomous disperse robot system. The morpho-configuration of a cluster robot system is mainly governed by the collision probability of its element modules that may be increased by introduction of telecommunication among the moving modules. This paper clarifies, by computer simulation, quantitatively the enhancement of module-connection rate, in the process of morpho-configuration, by a particular communication-module-action rule. The most important factor is found the orientation angle between two modules under communication when they start to approach each other. Moreover, we examine relationship between module speed and action. We realize critical speed for stability of cluster robot system.

View full abstract

We developed a home automation system called BlogAlpha that uses a home automation robot ApriAlpha^<TM> to integrate the legacy appliances in a home. It provides a blog interface to receive the user's request remotely in a natural language and show the state of the home. A robot works as intelligent glue that connects and automates legacy appliances, allowing users to introduce an intelligent home environment in their home at much lower cost. BlogAlpha uses the ontologies about the commodities in a home, the locations where these are placed and the tasks the robot can achieve.

View full abstract

It is difficult for the ordinary user to set up a complex robot system in the environment where two or more home robots exist. Then, we apply the functions of UPnP (Universal Plug and Play) to the robot system using the ORiN and the RAC for means of an information transmission base. As a result, we realized that the system becomes more flexible and the user doesn't have to set configuration. In this paper, we describe the solution and implementation of auto-configuration and user interface for robot service.

View full abstract

The paper describes the method for the teleoperation of the mobile robot through the wireless ad hoc sensor network in the environment such as disaster area. By using the sensor network, it is possible to communicate with the robot without using an existing communication infrastructure and use gathered information by the sensor network. The system configuration to use sensing information and routing on sensor network are discussed for the comfortable teleoperation. The teleoperation system by using sensor network device, mica2 mote, and omni-directional mobile robot are constructed and examined to confirm the utility of the system.

View full abstract

This research studies the estimation of the number of walking humans by using footsteps measured by the sensor network. We develop a new wireless communication module mounted the omni directional microphone. This module is able to digitally record the sound for about 16 seconds at the sampling frequency 8 kHz and communicate each module at 160Kbit per second. Using the sensor network constructed from these modules, and we measure the footstep of walking human. From the sensing data, we are able to estimate the number of walking humans around a module, walking pattern and walking direction. Furthermore the result implies the capability to estimate the motion of individual walking human by combining each node's data.

View full abstract

I have been trying to induce a biped robot with medium walking speed by using neural oscillators, based on the biological inspired system for a Quadruped walking robot "Tekken". I design the mechanical system and the neural system consisting of neural oscillators, responses and PD-controller. In this paper, I show a new inverted pendulum type biped robot. The biped robot is successful in simple dynamic walking on the flat terrain. My goal is developing a biped robot with high energy efficiency and high adaptability to environment.

View full abstract

We have already created a robot "Runbot2C" like a cat which succeeded walking and running in three dimensional planes. The Runbot2C have achieved average speed 1.14 [m/s] with using bounce gait. In this research, it proposed the method of controlling the gallop running. Moreover, it was confirmed to run by smoothly switching the gallop running from the bounce running.

View full abstract

Several types of running robots have already been studied and developed, In order to create a three-dimensional running robot, we study a control method called Variable Constraint Control which can realize various gaits. The motion pattern for the dinosaur-type robot is easily switched by changing the parameters of a differential equation. In recent years, FPGA (Field Programmable Gate Array) is used for control of several systems. FPGA can implement CPU, memory, digital filter, signal generator, and comparator all-in-one package. In this paper, we try realization of implementation control system based on FPGA in consider of a control method for the dinosaur type robot.

View full abstract

This paper discusses the possibility to realize both of walking and running by a single robot, which is made by the combination mechanism of linkages and actuators. This linkage mechanism have actuators below the number of joint therefore it enables to reduce the weight of the robot This study aims at realizing the 4-legged locomotion by controlling the phase to manipulate the legs made by linkage mechanism. This paper firstly shows improvements of issues of prototype I that was developed last year. Next, we shows design of prototype II that was improved above issues. Prototype II has improvement drive unit that is able to changing locomotion while moving.

View full abstract

In this paper, an obstacle avoidance method is proposed for quadruped robots by using camera image and ultrasonic sensors. Single camera is used to detect the corner of rectangular obstacles by image processing and acquire the rough information of 3-D obstacle. Ultrasonic sensors are used to obtain the accurate information of the obstacle. Actions of quadruped robots are defined as a forward action, a striding action, a climbing-over action and a detour action. Action selector can select a set of these actions. The effectiveness of the proposed obstacle avoidance method is tested through some experiments.

View full abstract