Deep generative models, particularly diffusion and flow matching models, have recently shown remarkable potential in learning complex policies through imitation learning. However, the safety of generated motions remains overlooked, particularly in complex environments with inherent obstacles. In this work, we address this critical gap by proposing Potential Field-Guided Flow Matching Policy (PF2MP), a novel approach that simultaneously learns task policies and extracts obstacle-related information, represented as a potential field, from the same set of successful demonstrations. During inference, PF2MP modulates the flow matching vector field via the learned potential field, enabling safe motion generation. By leveraging these complementary fields, our approach achieves improved safety without compromising task success across diverse environments, such as navigation tasks and robotic manipulation scenarios. We evaluate PF2MP in both simulation and real-world settings, demonstrating its effectiveness in task space and joint space control. Experimental results demonstrate that PF2MP enhances safety, achieving a significant reduction of collisions compared to baseline policies. This work paves the way for safer motion generation in unstructured and obstacle-rich environments.

In the current era of AI technology, where systems increasingly rely on big data to process vast amounts of societal information, efficient methods for integrating and utilizing diverse datasets are essential. This article presents a novel approach for transforming the feature space of different datasets through singular value decomposition (SVD) to extract common and hidden features as using the prior domain knowledge. Specifically, we apply this method to two datasets: 1) one related to physical and cognitive frailty in the elderly; and 2) another focusing on identifyingIKIGAI(happiness, self-efficacy, and sense of contribution) in volunteer staff of a civic health promotion activity. Both datasets consist of multiple sub-datasets measured using different modalities, such as facial expressions, sound, activity, and heart rates. By defining feature extraction methods for each subdataset, we compare and integrate the overlapping data. The results demonstrated that our method could effectively preserve common characteristics across different data types, offering a more interpretable solution than traditional dimensionality reduction methods based on linear and nonlinear transformation. This approach has significant implications for data integration in multidisciplinary fields and opens the door for future applications to a wide range of datasets.

Industrial robots require force controllability and impact resistance to ensure safe physical interactions. An electro-hydrostaic actuator (EHA) is expected to be suitable for such applications because it has high backdrivability which improve both force controllability at contact and impact resistance. However, EHAs had been rarely used in multi-axes robotic systems. The previous works validated the force controllability of the EHA-driven robot Hydra. However, the impact resistance of an EHA-driven robot is still unclear. In order to evaluate the impact resistance of the high-power EHA-driven robot, we developed high-pressure EHAs employing ceramics as rigid material to reduce internal leakage, and developed the EHA-driven 6-DOF robot arm Hydracer as the platform for the evaluation. This paper describes the mechanism of Hydracer especially on the base 3-DOF mechanism, and conducts the backdrivability measurement and the impact resistance evaluation.

This article discusses a compliance optimization approach that satisfies positive definiteness. Physical human–robot interactions are an important topic in robotics, for which force or compliance control is a key technology. Operational space control (OSC) is one of the most common approaches for robot force control with redundant degrees of freedom. By linearizing OSC, we can derive joint stiffness and viscosity matrices equivalent to the OSC. For an appropriate control, it is important that these matrices are positive definite. However, the stiffness matrix equivalent to the OSC is not always positive definite. In this case, a high kinetic energy is required, which is a problem in terms of the control performance. Therefore, the control performance can be improved by explicitly considering the positive definiteness of the stiffness or compliance. In this article, the authors derive a dynamically consistent compliance formulation and propose a compliance optimization that satisfies positive definiteness. The space of the symmetric positive definite matrix is a Riemannian manifold. We show that minimizing the Riemaniann geodesic distance results in a better performance compared with using OSC. The proposed method is validated via forward dynamics simulations and experiments using a hydrostatically driven humanoid Hydra.

6-DOF robot manipulator Hydracer was developed to gain high output torque and high backdrivability by adopting electro-hydrostatic actuators, however control of overall system of Hydracer is not yet conducted. To achieve flexible force control of Hydracer, we worked on the system improvements: enhancement of reliability of ceramics components, reduction of internal leakage by considering the property of hydraulic oil, and the identification of inertial parameters to improve its controllability. By using identified parameters, flexible force control of Hydracer by zero-torque control with gravity compensation was realized which reveals the potential of safe human-robot interaction.

Dynamic environments are challenging for visual SLAM since the moving objects occlude the static environment features and lead to wrong camera motion estimation. In this paper, we present a novel dense RGB-D SLAM solution that simultaneously accomplishes the dynamic/static segmentation and camera ego-motion estimation as well as the static background reconstructions. Our novelty is using optical flow residuals to highlight the dynamic semantics in the RGB-D point clouds and provide more accurate and efficient dynamic/static segmentation for camera tracking and background reconstruction. The dense reconstruction results on public datasets and real dynamic scenes indicate that the proposed approach achieved accurate and efficient performances in both dynamic and static environments compared to state-of-the-art approaches.

This paper discusses the application of video motion capturing technology (VMocap) to a competitive team sports game. The setting introduces a specific set of constraints: large scale markerless motion capturing, big recording volume, transmitting and processing gigabytes of data, operation without interfering with players or distracting spectators and staff, etc... In this paper, we present how we tackled and successfully solved all of these constraints. That enabled us to analyze the sportsmen without any intrusions, while giving their peak performance, hence opening a new field for Mocap application. International volleyball game was recorded in full length with the described system. During the course of the event, we compressed 54TB of raw image data real-time, capturing 6 hours of high framerate video per camera, without disturbing any of the game operations. Using the data, we were able to reconstruct the motion, muscle activity and behavior of the athletes present on the court.

We present SplitFusion, a novel dense RGB-D SLAM framework that simultaneously performs tracking and dense reconstruction for both rigid and non-rigid components of the scene. SplitFusion first adopts deep learning based semantic instant segmentation technique to split the scene into rigid or non-rigid surfaces. The split surfaces are independently tracked via rigid or non-rigid ICP and reconstructed through incremental depth map fusion. Experimental results show that the proposed approach can provide not only accurate environment maps but also well-reconstructed non-rigid targets, e.g., the moving humans.

This paper describes a stable knee-stretched walking of a humanoid by the resolved viscoelasticity control (RVC). The RVC method resolves multiple viscoelasticities in task-space, including the center of mass viscoelasticity for balancing, into joint-space viscoelasticity. Although a robust and compliant motion was achieved by the RVC method in previous studies, the conventional knee-bent posture to avoid the kinematic singularity suffered large knee joint torque. In this study, we propose an extension of the RVC capable of the kinematic singularity. We demonstrate through simulations and experiments that the RVC method considering the singularity achieves a stable and human-like walking, reducing the knee joint torque and improving the energy efficiency.

It is still an open problem to develop a reduced order model of bipedal walking that closely represents the complex dynamics of humanoid robots. In this paper, we propose control methodologies, removing both the constant CoM height constraint and the constant centroidal angular momentum constraint. We define a capturability criterion. and propose an enhanced intrinsically stable model predict control to fulfill this new capturability criterion. Then the angular momentum can be controlled. The results of simulations using humanoid robot HRP-4 show the proposed methods can improve the stability of bipedal locomotion on uneven terrains.

In this paper, we investigate the relationship between the depth perception and an approaching motion toward an object. We propose the depth perception dependency test, which is the combination of tests of a motion and depth perception based on the situation that an object is placed on the human's pathway. Firstly, as the motion test, we set the ball approach motion and asked elderly participants to approach and contact a ball by their foot, because this motion is easy to measure and requires localization skill and motion planning skill. The swing/support legs positions at the toeoff time of the swing leg just before contacting the ball was analyzed. Secondly, as the depth perception test, the pseudo3D image test was proposed. The coordinate transformation model for the calculation of the depth perception ability was also proposed. Through the proposed test and the proposed model, it was clarified that the participants who are regarded as perceiving the strong visual illusion perceive the objects closer than its real position, and their swing leg toe off positions in the ball approaching motion were significantly farther than other participants. Thus, it can be concluded that there was a relationship between the depth perception and the approaching motion that is thought to be a higher risk of stumbles.

The paper aims to develop a methodology of transferring the knowledge obtained from the experiments of laboratory animals to human musculoskeletal system. To achieve the goal, we propose a method for estimating the homologous posture of the mammalian skeletal system corresponding to the human body posture. We hypothesize the homology of bone geometry between mammalian species implies that of biomechanical functions. The method relies on this homology and determines the homologous postures according to the anatomical landmarks of bone geometry. This paper shows the results of the analysis on homologous postures between the human and mouse skeletal models to validate our hypothesis. A pilot study also introduces comparison of mechanical functions between the two models by using the homologous postures.

This paper discusses video motion capture, namely, 3D reconstruction of human motion from multi-camera images. After the Part Confidence Maps are computed from each camera image, the proposed spatiotemporal filter is applied to deliver the human motion data with accuracy and smoothness for human motion analysis. The spatiotemporal filter uses the human skeleton and mixes temporal smoothing in two-time inverse kinematics computations. The experimental results show that the mean per joint position error was 26.1mm for regular motions and 38.8mm for inverted motions.

Motor functions of the biological system has been forged through 4 billion years evolution. From a neurorobotics view, it is important not only to know how well it works, but also how it fails. To quantitatively describe early onset symptoms of a neurodegenerative disease, we analyzed phenotypes of genetically engineered Huntington disease (HD) model mice, which reveal progressive impaired motor functions. We devised a simple yet sensitive paradigm called the crystalized motion profile (CMP), by which we successfully detected subtle difference between normal and abnormal mice in terms of whole-body level motor coordination. Our long-term objective is to remodel human mind and body to regain impaired motor and cognitive functions with ageing. To do so, we are developing a soft neurorobotic suit that provides integrated cognitive and physical interventions to users. Our analysis on the HD model mice is important as the first step to bridge between molecular mechanisms (altered genetic code) and the macroscopic neuro-musculoskeletal model. With this, we can extrapolate from knowledge of non-human mammals to human to derive the remodeling.

In this paper, we propose a motion model that focuses on the discriminative parts of the human body related to target motions to classify human motions into specific categories, and apply this model to multi-class daily motion classifications. We extend this model to a motion recognition system which generates multiple sentences associated with human motions. The motion model is evaluated with the following four datasets acquired by a Kinect sensor or multiple infrared cameras in a motion capture studio: UCF-kinect; UT-kinect; HDM05-mocap; and YNL-mocap. We also evaluate the sentences generated from the dataset of motion and language pairs. The experimental results indicate that the motion model improves classification accuracy and our approach is better than other state-of-the-art methods for specific datasets, including human–object interactions with variations in the duration of motions, such as daily human motions. We achieve a classification rate of 81.1% for multi-class daily motion classifications in a non cross-subject setting. Additionally, the sentences generated by the motion recognition system are semantically and syntactically appropriate for the description of the target motion, which may lead to human–robot interaction using natural language.

For heavy duty tasks which are needed in field or rough terrain, we developed a hydrostatically actuated anthropomorphic hand. The hand is specifically designed for the humanoid robot HYDRA, whose 40 joints are driven by back-drivable electro-hydrostatic actuators (EHA). Each designed hand has four fingers with a total of five DOF. Each finger has three joints underactuated by one tendon. Opposition/reposition of the thumb joint is also driven by one tendon. The five tendons are pulled by a miniature linear cluster EHA mounted in the forearm. The cluster EHA consists of a light weight tie-rod cylinder cluster with five pistons, and five low friction trochoid pumps with a crescent separator. Its 300 N nominal tension generates 1.5 Nm joint torque on each of the finger joints. Design of the cylinder and pump, with results of evaluation experiments is shown in this paper. Forearm structure with a mechanism to measure the tendon tension, low friction tendon routing in the forearm, low friction wire guiding link for the wrist, and parallel link wrist driving mechanism with two EHAs are also described.

We propose a method for checking and enforcing multicontact stability based on the zero-tilting moment point (ZMP). The key to our development is the generalization of ZMP support areas to take into account: 1) frictional constraints and 2) multiple noncoplanar contacts. We introduce and investigate two kinds of ZMP support areas. First, we characterize and provide a fast geometric construction for the support area generated by valid contact forces, with no other constraint on the robot motion. We call this set the full support area. Next, we consider the control of humanoid robots by using the linear pendulum mode (LPM). We observe that the constraints stemming from the LPM induce a shrinking of the support area, even for walking on horizontal floors. We propose an algorithm to compute the new area, which we call the pendular support area. We show that, in the LPM, having the ZMP in the pendular support area is a necessary and sufficient condition for contact stability. Based on these developments, we implement a whole-body controller and generate feasible multicontact motions where an HRP-4 humanoid locomotes in challenging multicontact scenarios.

Humanoids locomote by making and breaking contacts with their environment. Thus, a crucial question for them is to anticipate whether a contact will hold or break under effort. For rigid surface contacts, existing methods usually consider several point-contact forces, which has some drawbacks due to the underlying redundancy. We derive a criterion, the Contact Wrench Cone (CWC), which is equivalent to any number of applied forces on the contact surface, and for which we provide a closed-form formula. It turns out that the CWC can be decomposed into three conditions: (i) Coulomb friction on the resultant force, (ii) CoP inside the support area, and (iii) upper and lower bounds on the yaw torque. While the first two are well-known, the third one is novel. It can, for instance, be used to prevent the undesired foot yaws observed in biped locomotion. We show that our formula yields simpler and faster computations than existing approaches for humanoid motions in single support, and assess its validity in the OpenHRP simulator.

Gesture recognition is used for many practical applications such as human-robot interaction, medical rehabilitation and sign language. In this paper, we apply a hybrid generative-discriminative approach by using the Fisher Vector to improve the recognition performance. The strategy is to merge the generative approach of Hidden Markov Model dealing with spatio-temporal motion data with the discriminative approach of Support Vector Machine focusing on the classification task. The motion segments are encoded into HMMs, and each segment is converted to FV, whose elements can be obtained as the derivative of the probability of the segment being generated by the HMMs with respect to their parameters. SVM is subsequently trained by the FVs. An input gesture can be classified to corresponding gesture category by SVM. In the experiments, we test our approach by comparing three HMM chain models and four categorization methods on dataset provided by the ChaLearn Looking at People Challenge 2014 (LAP 2014). The results show that similar gesture patterns are clustered closely in several categories. Our approach based left-to-right HMMs outperforms other gesture recognition methods. More specifically, the hybrid generative-discriminative approach overcomes the standard HMM approach and the generative kernel approach overcomes the generative embedding approach. For these results, our approach is effective to improve the recognition performance.

This paper discusses an online dynamic motion generation scheme for nonprehensile object manipulation by using a set of predefined motions and a trajectory deformation algorithm capable of incorporating positional and velocity boundary constraints. By creating optimal trajectories offline and deforming them online, computational complexity during execution is reduced considerably. As tight convex hulls of the deformed trajectories can be found, possible obstacles or workspace boundaries can be circumnavigated precisely without collision. The approach is verified through experiments on an inclined planar air-table for volleyball scenario using two 3-DoF robots.

We propose a new approach to deform robot trajectories based on affine transformations. At the heart of our approach is the concept of affine invariance: Trajectories are deformed in order to avoid unexpected obstacles or to achieve new objectives but, at the same time, certain definite features of the original motions are preserved. Such features include, for instance, trajectory smoothness, periodicity, affine velocity, or more generally, all affine-invariant features, which are of particular importance in human-centered applications. Furthermore, this approach enables one to “convert” the constraints and optimization objectives regarding the deformed trajectory into constraints and optimization objectives regarding the matrix of the deformation in a natural way, making constraints satisfaction and optimization substantially easier and faster in many cases. As illustration, we present an application to the transfer of human movements to humanoid robots while preserving equiaffine velocity, a well-established invariant of human hand movements. Building on the presented affine deformation framework, we finally revisit the concept of trajectory redundancy from the viewpoint of group theory.

In this paper, we propose an approach for learning task specifications automatically, by observing human demonstrations. Using this approach allows a robot to combine representations of individual actions to achieve a high-level goal. We hypothesize that task specifications consist of variables that present a pattern of change that is invariant across demonstrations. We identify these specifications at different stages of task completion. Changes in task constraints allow us to identify transitions in the task description and to segment them into subtasks. We extract the following task-space constraints: 1) the reference frame in which to express the task variables; 2) the variable of interest at each time step, position, or force at the end effector; and 3) a factor that can modulate the contribution of force and position in a hybrid impedance controller. The approach was validated on a seven-degree-of-freedom Kuka arm, performing two different tasks: grating vegetables and extracting a battery from a charging stand.

The panorama of probabilistic completeness results for kinodynamic planners is still confusing. Most existing completeness proofs require strong assumptions that are difficult, if not impossible, to verify in practice. To make completeness results more useful, it is thus sensible to establish a classification of the various types of constraints and planning methods, and then attack each class with specific proofs and hypotheses that can be verified in practice. We propose such a classification, and provide a proof of probabilistic completeness for an important class of planners, namely those whose steering method is based on the interpolation of system trajectories in the state space. We also provide design guidelines for the interpolation function and discuss two criteria arising from our analysis: local boundedness and acceleration compliance.

This paper presents a novel robot pose measure for human movement imitation based entirely on the Euclidean distance information between any two links of a robot and any link and object in the robot's environment in a Cartesian task space. A Hidden Markov Model is used to encode the spatio-temporal information of multiple demonstrations. In combination with Gaussian Mixture Regression for extracting the important task properties, feasible full-body motion adaption can be achieved. The method is suited for use with a humanoid robot by considering additional constraints like balance control and collision avoidance. In order to tackle modeling errors occurring due to the human movement demonstration and the robotic reproduction, a manipulability based weighting scheme is proposed. Complexity reduction of the otherwise redundant pose measure is performed based upon a mechanical analogy of an interconnected spring system. Experiments are conducted using a HRP-4 robot and display the applicability of the presented methods for robotic full-body motion imitation tasks.

Underactuated hands usually have high adaptability in power grasping but they are limited in pinching task with fingertip. In this paper, we propose the design of a tendon-driven underactuated hand that is capable of fingertip pinching by utilizing our proposed coupling mechanism. To reduce the friction resulting from tendon routing, we introduce the carpal tunnel mechanism that replace all sliding-contact tendon routing with the pulley system allocated in 3-dimensional space. The prototype of 11-DOF anthropomorphic robot hand is fabricated using rapid prototyping. Experiments are done to prove the effectiveness of our proposed coupling mechanism and low-friction tendon-driven system for underactuated robot hand.

This paper presents an incremental sampling-based approach for trajectory imitation in cluttered environments using the RRT* algorithm. Inspired by the discrete Laplace-Beltrami operator the underlying distance metric is based upon the difference from a reference trajectory through a quadratic distance term incorporating velocity and acceleration deviations along the trajectory. Mathematically-backed approximations in combination with a task-space bias make it possible to use standard nearest neighbor methods in task space when expanding the RRT*-tree. It is shown that metric-consistent biases considerably increase the convergence speed. The proposed approach is validated in simulations in a 2D environment and in experiments using a HRP-4 humanoid robot.

This paper proposes a new torque sensing method, Cr-N alloy strain sensitive thin-film based torque sensors and distributed joint torque servo systems that enable human support robots to have capabilities to make physical interaction in adaptive and safe operation tasks. Stiffer torque sensing with stable and high-resolution sensing to meet practical level of the developed torque sensors have been achieved. We have developed the joint torque control based 4-DOF arm models in order to verify practical effectiveness of the proposed joint torque servo systems. We also demonstrated the joint torque control based bilateral master slave system exploring future applications.

In this paper, we demonstrate that a musculoskeletal model with an anatomically-inspired abdomen model can estimate human-like abdominal muscle tension, in contrast to previous musculoskeletal models or models based on joint elasticity. We first model the abdominal structure based on anatomical knowledge. The anatomically-inspired abdomen model consists of a Tendinous Intersection model that connects the muscle bellies of Rectus Abdominis and represents its pathway, a Rectus Abdominis model that represents the muscle contraction part, and a balloon-type abdominal cavity model that represents the volumetric abdominal cavity. These models compute the intra-abdominal pressure and its effect on the lumbar vertebrae joint torques. We use these models with our musculoskeletal model to estimate the abdominal muscle tension during sitting up motion with optical motion capture, inverse kinematics, and inverse dynamics computation. Our computational results show that our abdomen model estimates the muscle tension with a waveform similar to the muscle activity measured by electromyography, while the previous musculoskeletal model and the joint elasticity model estimate obviously different muscle tension patterns. These results imply that the intra-abdominal pressure is critical in estimating the abdominal muscle tension rather than the inertial properties and the joint elasticity.

Generating a physically feasible pair of motion pattern and force pattern for walking motions of a humanoid robot can be analyzed as a two point boundary value problem. The final boundary condition plays a critical role as the state trajectory depends on it and the remaining time to the goal. In this paper, the effect of the boundary condition is exploited to compensate disturbances during the walking motion and two strategies to modify it are proposed. The first one is based on the neutral point condition and it prioritizes stabilization in one step. The second one is based on an optimal formulation which can include future steps in the cost function. The unilateral constraint on the reaction force and the kinematic workspace are taken into account to compare both reactive control methods.

Motion recognition is an essential technology for social robots in various environments such as homes, offices and shopping center, where the robots are expected to understand human behavior and interact with them. In this paper, we present a system composed of three models: motion language model, natural language model and integration inference model, and achieved to generate sentences from motions using large high-order N-grams. We confirmed not only that using higher-order N-grams improves precision in generating long sentences but also that the computational complexity of the proposed system is almost the same as our previous one. In addition, we improved the precision by aligning the graph structure representing generated sentences into confusion network form. This means that simplifying and compacting word sequences affect the precision of sentence generation.

This paper introduces Locally Weighted Least Squares Policy Iteration for learning approximate optimal control in settings where models of the dynamics and cost function are either unavailable or hard to obtain. Building on recent advances in Least Squares Temporal Difference Learning, the proposed approach is able to learn from data collected from interactions with a system, in order to build a global control policy based on localised models of the state-action value function. Evaluations are reported characterising learning performance for non-linear control problems including an under-powered pendulum swing-up task, and a robotic door-opening problem under different dynamical conditions.

The importance of force measurement and back-drivability in realizing force sensitive actuator is widely acknowledged. There are studies on fidelity of torque sensors and backdrivability individually, but limited study are made on investigating effect of torque fidelity and backdrivability on force sensitivity of the actuation system. In this paper, we developed backdrivable electro-hydrostatic actuator equipped with torque sensor to analyze the effect of torque fidelity and backdrivability on force sensitive control system. We implemented friction compensation controller and evaluated force sensitivity of the actuator by residual friction torque after the friction compensation. Method using pressure sensor and torque sensor were compared. Effect of backdrivability was performed by comparing friction torque of Harmonic Drive joint and joint with developed actuator.

This study proposes a model that combines a realistically scaled neural network made up of pools of spiking neurons, with a musculoskeletal model of the human arm. We used evidence from literature to design topological pools of spinal neurons and their synaptic connections. The spiking output of the motor neuron pools were used as the command signals that generated motor unit forces, and drove joint motion. Feedback information from muscle spindles was relayed to the neural network via monosynaptic and disynaptic pathways. Participant-specific parameters of the combined neuromusculoskeletal (NMS) system were then identified from recorded experimental data. The identified NMS model was used to simulate the arm stretch reflex and the results were validated by comparison to an independent recorded dataset. The models and methodology proposed in this study show that large and complex neural systems can be identified in conjunction with the musculoskeletal systems that they control. This additional layer of detail in NMS models has important relevance to the research communities related to rehabilitation robotics and human movement analysis.

Musculoskeletal models are effective tools for understanding living systems. To ensure proper model function, they must be checked against the literature or specimens. Existing checking methods require cadaver experimentation, highly knowledgeable medical personnel, and/or significant time. In this paper, we propose a quick and efficient method, called functional consistency checking, for use when these resources are not available. This method uses the literature to define a set of mathematical constraints, custom inverse dynamics software to interact with the model and its Jacobian in realtime and then evaluates the models consistency with these constraints. The method's usefulness will be demonstrated by constructing a human hand prototype, performing functional consistency checking, and then comparing the original to the output using data from a pianist motion capture.

It is widely acknowledged that the actuator's intrinsic backdrivability is important in realizing a force sensitive behavior. It is desirable to realize such actuator with electric motor that is advantageous from power-to-weight ratio and controllability point of view. Electro-Hydrostatic Actuator is a type of hydraulic actuators that can realize high backdrivability by reducing transmission friction and by providing dynamics decoupling with an implicit serial damper. To farther enhance the backdrivability of a EHA, a pump with minimum static and Coulomb friction is necessary. In this paper, we introduce an EHA with viscous screw pump that minimizes static and Coulomb friction by eliminating the mechanical contact between pump components. Viscous screw pumps also have the advantage that there is no pulsation in pressure due to the continuity of the force transmission from the rotor to the fluid. The property of the actuator, including pulsation performance and impedance control performance were evaluated on a prototype of EHA with a viscous screw pump.

Our contribution in this article is twofold. First, we identify the regularity properties of the trajectories of planar wheeled mobile robots. By regularity properties of a trajectory we mean whether this trajectory, or a function computed from it, belongs to a certain class Cn(the class of functions that are differentiable n times with a continuous nthderivative). We show that, under some generic assumptions about the rotation and steering velocities of the wheels, any non-degenerate wheeled robot belongs to one of the two following classes. Class I comprises those robots whose admissible trajectories in the plane are C1and piecewise C2; and class II comprises those robots whose admissible trajectories are C1, piecewise C2and, in addition, curvature-continuous. Second, based on this characterization, we derive new feedback control and gap-filling algorithms for wheeled mobile robots using the recently-developed affine trajectory deformation framework.

The language is a symbolic system unique to human being. The acquisition of language, which has its meanings in the real world, is important for robots to understand the environment and communicate with us in our daily life. This paper proposes a novel approach to establish a fundamental framework for the robots which can understand language through their whole body motions. The proposed framework is composed of three modules: “motion symbol”, “motion language model”, and “natural language model”. In the motion symbol module, motion data are symbolized by Hidden Markov Models (HMMs). Each HMM represents abstract motion patterns. Then the HMMs are defined as motion symbols. The motion language model is stochastically designed for links between motion symbols and words. This model consists of three layers of motion symbols, latent states and words. The connections between the motion symbol and the latent state, and between the latent state and the words are denoted by two kinds of probabilities respectively. One connection is represented by the probability that the motion symbol generates the latent state, and the other connection is represented by the probability that the latent state generates the word. Therefore, the motion language model can connect the motion symbols to the words through the latent state. The natural language model stochastically represents sequences of words. In this paper, a bigram, which is a special case of N-gram model, is adopted as the natural language model. This model has the words as nodes and transitions between two words as edges. Therefore sentence structure is expressed as transitions among words. The integration of the motion language model and natural language model can be implemented by the search computation for sentences corresponding to motions and for motions corresponding to sentences. Especially, the usage of the bigram as the natural language model provides a simple search computation so that appropriate and fast bidirectional computation between the motions and language can be achieved. Our approach makes it possible for humanoid robots not only to interpret motions as sentences but also to generate motions from sentences. The tests by using various motions and words validate our framework for the language acquisition of humanoid robots.

This paper presents a fast inverse kinematics (IK) algorithm. In recent years, the robotics computation theory is often applied for detailed and complex multi-body systems. However, the computational complexity of IK is too high to be implemented in large DOF systems. IK of multi-body system is often formulated as nonlinear optimization to minimize the residuals from the references. It usually requires the computation of the gradient vector of the evaluation function. In the method, the computation of the gradient is decomposed into two parts. First, the residuals are considered as external forces and are distributed to each link. Then, the gradient can be computed from static equilibrium by the recursive Newton-Euler algorithm. In addition with the efficient direction search algorithms of nonlinear programing, the computation complexity of IK can be dramatically reduced. The results of numerical evaluation using a large-DOF manipulator and a human musculoskeletal model are shown.

In this paper we present a framework for generating walking motions for a humanoid robot and how to adapt the trajectory to handle disturbances during the execution of the trajectory. Based on the simplified dynamics of a mass concentrated model, we generate a physically consistent motion that depends on the current CoM state space of the robot. If the disturbance is low, it is possible to adapt the trajectory in order to reach the desired next step position. In the case of strong disturbances, we show how to modify the final goal under a reactive stepping framework. Simulations and hardware experiments with the full-size humanoid HRP-4 show the validity of the proposed scheme.

To identify the joint parameters (e.g. the position of the joint center for a spherical joint, the position and the orientation of the joint axis for a revolute joint, etc.) from motion capture data, existing provably-correct algorithms require that at least three markers be attached to either of the two links adjacent to the joint. However, as shown in this article, it turns out that the identification of the joint parameters requires, for most types of joints, strictly less than three markers on any link. More precisely, we prove the structural identifiability of joint parameters in the following cases: (a) a spherical joint with two markers attached to each of the two adjacent links; (b) a revolute joint with two markers attached to one of the two links, and one marker attached to the other. We provide a practical algorithm to do the identification in case (a). Finally, we show that identification cannot be achieved with strictly fewer markers than listed in (a) and (b).

In this paper, we propose the inverse kinematics method based on high-order moment features and their Jacobian matrices, which can use an arbitrary information source about the shape of the targeted kinematic chain as reference input. The method is especially useful to generate the motion of humanoid robots and human figures, and we can generate the whole body pose from a set of 3D markers or pixels of 2D images, without labeling each feature point. The moment features are computed from various types of quantities, for example, geometric points, mass density, pixel images, and the probability labeled as a specific link, the fact of which shows generality and versatility of the method. Some results of motion of a human figure from the label-less 3D markers and the 2D images are illustrated.

The mass parameters of the human body segments are important when studying motion dynamics and the in vivo method to obtain accurate parameters is required in biomechanics studies and for some medical applications. In our previous works, we proposed the method to identify inertial parameters of human body segments in real-time during measurement of motion. However, some obtained parameters are not physically consistent; some masses are negative and inertia tensor matrices are not positive definite. These parameters generate problems in the analysis and the simulation requiring physical consistency. In this paper, we propose the real-time identification method considering physical consistency.

Realization of rigid and sensitive torque sensor is one of the key factors for the success of robots. With the conventional detectors as strain gauges, poor S/N (signal to noise) ratio has been the limitation of torque sensor sensitivity. Torque Encoder uses a linear encoder as a detector and significantly enhanced the S/N ratio, and realized stiff and sensitive torque sensor. However, the crosstalk in torque measurement was still an open problem as in other sensing methods. In this paper, we analyzed the cause of the crosstalk and proposed the mechanism to suppress the crosstalk using selectively compliant suspension mechanism. Design methodology, implementation to minimize crosstalk with minimal sensitivity loss are presented. Evaluation on a prototype of the mechanism is carried out to show the effectiveness of the mechanism.

This video presents our recent research on the integration of physical human-robot interaction (pHRI) into imitation learning. First, a marker control approach for real time human motion imitation is shown. Secondly, physical coaching in addition to observational learning is applied for the incremental learning of motion primitives. Last, we extend imitation learning to learning pHRI which includes the establishment of intended physical contacts. The proposed methods were implemented and tested using the IRT humanoid robot and DLR's humanoid upper-body robot Justin.

This paper proposes an approach to construct a system which allows humanoid robots to recognize human behaviors and predict his or her future behaviors. The system consists of two modules : "motion symbol tree" and "motion symbol graph", Human demonstrator motion patterns are stored as motion symbols, which abstract the motion data by using Hidden Markov Models. The stored motion patterns are organized into a hierarchical tree structure, which represents the similarity among the motion patterns and provides abstracted motion patterns. The formed hierarchical structure is the motion symbol tree. Concatenated sequences of motion patterns are stochastically represented as transitions between the motion patterns by using an Ngram Model, and the causality among the human behaviors are extracted. This structure is the motion symbol graph. The behavioral hierarchy and transition model make it possible to predict human behaviors during observation and to generate sequences of motion patterns automatically while maintaining a natural motion stream, as if the system is a "crystal ball" to reflect future behaviors. The experiments demonstrate the validity of the proposed framework on a large scale motion data.

Captured human motion data are used so that humanoid robots or computer graphics (CG) characters can behave naturally. Because the motion capture system is expensive, and time-consuming process is needed to acquire motion data, technology that enables to reuse existing motion data efficiently is required. This paper proposes motion retrieval method with natural language word based on stochastic correlation between motion and language. We construct a space which has maximum correlation with motion pattern features and word features, and we use this space as search space for motion retrieval. Proto-symbol space, which represents the relationship of each symbolized motion patterns, is used as motion feature space. And as word feature, binary features are used which represent whether a word label is attached or not. Because the constructed search space has correlation between motion patterns and words, associative motion retrieval considering similarity of motion pattern or closeness of word meaning becomes possible. We validate proposed motion retrieval method by constructing motion database with captured human motion data.

In current biomechanics approach, the assumptions are commonly used in body-segment parameters and muscle strength parameters due to the difficulty in accessing those subject-specific values. Especially in the rehabilitation and sports science where each subject can easily have quite different anthropometry and muscle condition due to disease, age or training history, it would be important to identify those parameters to take benefits correctly from the recent advances in computational musculoskeletal modeling. In this paper, Mass Distribution Identification to improve the joint torque estimation and Muscle Strength Identification to improve the muscle force estimation were performed combined with previously proposed methods in muscle tension optimization. This first result highlights that the reliable muscle force estimation could be extracted after these identifications. The proposed framework toward subject-specific musculoskeletal modeling would contribute to a patient-oriented computational rehabilitation.

There are many researches about mobile robot with a varied wheel. Mobile robots or mobilities with wheels such like a car or bike have non-holonomic constraint and then it is difficult to move toward wheel's axis direction. Namely, it is impossible to move toward omni-direction due to the constraints by wheel's mechanical structure. Omni-directional movement means the mobility can move toward any direction at anytime. This paper adapted link-driven spherical wheel as an omnidirectional wheel and solved singularity problem caused by mechanical structure of it. We developed omni-directional mobilities( SO(2) and SO(3)) installing link-driven spherical wheels and experimented omni-directional movement with designed controllers.

The mass properties are important to control robot dynamics or study human dynamics. In our previous works, we proposed a method to identify inertial parameters of legged mechanisms from base-link dynamics, using generalized coordinates and external forces information. In this paper, we propose an identification method based on floating-base dynamics, when the system has no external force. Inertial parameters can be identified without force measurement, only from motion data. The method has been tested on two examples; a simple chain consisted of two links and the human body dynamics.

Although backdrivability is recognized as an important property of actuators, this term is often used without clear definition. In order to design mechanisms with advanced controllability, it is important to understand the fundamental mechanism of backdrivability. In this paper, we introduced idea of series elasticity and series dissipation of actuators. Based on this idea, total/output backdrivability and their fundamental properties are stated. EHA was shown to be series dissipative and it was confirmed from the model of the actuator. Utilizing the backdrivability of EHA, position based impedance control was implemented and evaluated. Application of this EHA in a robot hand is also reported.

When robots cooperate with humans it is necessary for robots to move safely on sudden impact. Joint torque sensing is vital for robots to realize safe behavior and enhance physical performance. Firstly, this paper describes a new torque sensor with linear encoders which demonstrates electro magnetic noise immunity and is unaffected temperature changes. Secondly, we propose a friction compensation method using a disturbance observer to improve the positioning accuracy. In addition, we describe a torque feedback control method which scales down the motor inertia and enhances the joint flexibility. Experimental results of the proposed controller are presented.

Human motion coordination is a long-standing research issue in biomechanics, and it should also have some implications for humanoid robot control.We have built a whole-body somatosensory reflex model based on our neuromusculoskeletal model and identified its parameters through non-invasive measurements and statistical analysis. Such models are crucial for analyzing and estimating signals in the nervous system. In this paper, we focus on signal transmission delay of the somatosensory reflex loop and investigate its relationship with the generalization capability of the reflex model. We obtain some sets of model parameters assuming different time delays using the data obtained from a stepping motion, and perform cross validations against stepping motions with different cycles as well as entirely different behaviors such as squat and jump. Interestingly, time delays close to the value expected from physiological properties show better cross-validation results than others. This result suggests that relatively simple reflex control can be generalized to multiple behaviors if the parameters are appropriate, and that robust control is possible even with large feedback delay.

Impedance and Admittance Control are two distinct implementations of the same control goal. It is well known that their stability and performance properties are complementary. In this paper, we present a hybrid system approach, which incorporates Impedance and Admittance Control as two extreme cases of one family of controllers. This approach allows to continuously switch and interpolate between Impedance and Admittance Control. We compare the basic stability and performance properties of the resulting controllers by means of an extensive case study of a one-dimensional system and present an experimental evaluation using the KUKA-DLR-lightweight arm.

This paper describes an novel approach towards linguistic processing for robots through integration of a motion language model and a natural language model. The motion language model works for association of words from motion symbols. The natural language model is one used for a morphological analysis, which has been developed in natural language community. The natural language model is optimized using a enormous amount of words. So this model is scalable architecture. The motion language model and the natural language model can be integrated since both models are represented graphically. The integration of the motion language model and the natural language model allows robots not only to interpret motion patterns as sentences but also to generate motions from sentences. This paper demonstrates the validity of our proposed framework even in the case that large-scale word corpus is needed processing through experiments of interpreting motion patterns as sentences and generating motion patterns from sentences.

Humanoid robots should be able to stand and walk in the presence of external disturbances. This paper addresses the robustness of a humanoid robot to unknown disturbances. Applying the maximal CPI set, it becomes possible to consider the physical constraint explicitly in the COG-ZMP inverted pendulum model control. In our previous research, the convergence speed of COG was improved by applying the switching control based on the maximal CPI set to the stabilization control assuming the contact region is constant. This paper presents updating calculation method of the maximal CPI set when the contact region changes, and the authors propose a falling avoidance control as an application of it. Detecting the stepping necessity based on the maximal CPI set enables to unify the upright position stabilization and stepping motion for falling avoidance. The validity of the proposed method is verified with experiments.

The identification method for industrial manipulators considering physical consistency such as positive definiteness of inertial parameters has been developed, however it has to solve the quadratic programming with the non-linear inequality constraints. In identifying the large DOF systems like humanoid robots, the converged solution is difficult to be obtained. In this paper, we propose the method to realize physical consistency and computational stability. As inertial parameters of each link are represented with a finite number of mass points, the constraints can be approximated by linear inequalities. We also propose to solve the optimization problem, which minimizes the errors both from measured data and the priori parameters extracted from the geometric model like CAD data. The method can estimate standard inertial parameters, which is a useful notation to be used for other applications.

Backdrivability is a keyword with rising importance, not only in humanoid robots, but also to wearable robots. Power assist robots, namely exoskeletons are expected to provide mobility and independence to elderly and people with disability. In such robots, without backdrivability, error between human and robot motion can be painful; sometimes dangerous. In this research, we apply a type of hydraulic actuator, specially designed to realize backdrivability, to exoskeletal robot to improve backdrivability. We present the basic methodology to apply such hydraulic actuator to wearable robots. We developed prototype of knee joint power assist exoskeleton and performed fundamental tests to verify the design method validity.

This paper proposes a novel approach for extracting a model of movement primitives and their sequential relationships during online observation of human motion. In the proposed approach, movement primitives, modeled as hidden Markov models, are autonomously segmented and learned incrementally during observation. At the same time, a higher abstraction level hidden Markov model is also learned, encapsulating the relationship between the movement primitives. For the higher level model, each hidden state represents a motion primitive, and the observation function is based on the likelihood that the observed data is generated by the motion primitive model. An approach for incremental training of the higher order model during online observation is developed. The approach is validated on a dataset of continuous movement data.

This paper proposes an approach to hierarchy formation of human behaviors, extraction of the behavioral transitions, and their application to prediction and automatic generation of behaviors. Human demonstrator motion patterns are stored as motion symbols, which abstract the motion data by using Hidden Markov Models. The stored motion patterns are organized into a hierarchical tree structure, which represents the similarity among the motion patterns and provides abstracted motion patterns. Concatenated sequences of motion patterns are stochastically represented as transitions between the abstracted motion patterns by using an Ngram Model, and the transitional relationships of the human behaviors are extracted. The behavioral hierarchy and transition model make it possible to predict human behaviors during observation and to generate sequences of motion patterns automatically while maintaining a natural motion stream, as if the system is a “crystal ball” to reflect future behaviors. The experiments validates the proposed framework by using a developed visualization system, which shows the demonstrator or the operator the established hierarchical tree and the transition network of the motion patterns, predicted behaviors and generated sequences of the motion patterns.

Robots must have similar mechanical impedance characteristics to humans in order to make safe and efficient contact. This impedance requirement applies not only to the surface but also to the actuation mechanisms. The objective of this research is to develop inherently flexible actuator by realizing backdrivability. A class of hydraulic actuation called electro-hydrostatic actuator was applied to knee joint in humanoid robots to satisfy flexibility and large torque output simultaneously. This paper explains the methodology of performance evaluation of actuators and design concept of joint mechanism. Mathematical model of electro-hydrostatic transmission is also presented. Evaluation of backdrivability, inertia modification control, and compliance control of developed mechanism are performed.

This paper proposes a novel approach for motion primitive segmentation from continuous full body human motion captured on monocular video. The proposed approach does not require a kinematic model of the person, nor any markers on the body. Instead, optical flow computed directly in the image plane is used to estimate the location of segment points. The approach is based on detecting tracking features in the image based on the Shi and Thomasi algorithm [1]. The optical flow at each feature point is then estimated using the Lucas Kanade Pyramidal Optical Flow estimation algorithm [2]. The feature points are clustered and tracked on-line to find regions of the image with coherent movement. The appearance and disappearance of these coherent clusters indicates the start and end points of motion primitive segments. The algorithm performance is validated on full body motion video sequences, and compared to a joint-angle, motion capture based approach. The results show that the segmentation performance is comparable to the motion capture based approach, while using much simpler hardware and at a lower computational effort.

In this paper, mimetic communication is extended to human-robot interaction tasks, in which physical contact transitions must be handled. The mimetic communication consists of imitation learning for learning low level motion primitives and a higher level interaction learning stage in which also the information about the human-robot contacts is included. For the imitation learning, Cartesian marker data from a motion capture system is used. A modification of the low level marker trajectory following algorithm is presented, which allows to reshape the trajectory of the motion primitive in accordance with the human hand motion in real-time. Moreover, for performing safe contact motion, an appropriate impedance controller is integrated into the setting. All the presented concepts are evaluated in experiments with a humanoid robot.

This paper describes the linguistic model based on symbolization of motion patterns for humanoid robots. The model consists of two kinds of stochastic models : the motion language model and the natural language model. The motion language model stochastically connects the symbols of motion patterns to the morpheme words through the latent states which represent the underlying linguistic structure such as semantic contents. The natural language model represents the dynamics of the word classes. The motion language model and the natural language model correspond to semantics and syntax respectively. The integration of the motion language model and the natural language model allows robots not only to linguistically interpret the motion patterns as sentences but also to generate the motions from the sentences. The two kinds of linguistic processes of the interpretation and the generation can be obtained by solving search problems: search for a sequence of morpheme words and a symbol of motion pattern. The proposed approach to interpretation of motion patterns as sentences and generation of motion patterns from the sentences through integration of the motion language model and the natural language model is validated by the experiment on the human behavioral data.

Identification results dramatically depend on the excitation properties of the motion used to sample the identification model. Strategies to define persistent exciting trajectories have been developed for manipulator robots with few DOF. However they can not easily be extended to humanoid systems and humans due to the important number of DOF; and empirical knowledge is often used to generate and select persistent exciting motions. In this paper we propose a method to choose persistent exciting motions from an existing dataset in order to optimize both the identification results and the computation time. This method is based on the use of the identification model of legged systems obtained from the base-link equations. Instead of using well-established consideration on the condition number of the regressor matrix, the method uses a decomposition of the regressor into elementary sub-regressors and the computation of the condition number for each. A selection rule is then proposed. The overall method is experimentally tested to identify the human body inertial parameters using a data-set of 40 motions. Comparative results obtained from different combinations of motions are given.

Contact-free estimation of the human somatosensory information is an essential skill for robots working in daily environments. The main objective of this paper is to develop a method for estimating muscle tensions without any sensors attached to the body. Muscle tension is an important information for evaluating physical load during motions. Existing approaches utilizing optimization techniques and/or electromyography (EMG) signals are not appropriate due to lack of physiological validity or usage of electrodes. In this paper, we propose to use a database of muscle tension distribution for obtaining physiologically realistic muscle tensions only from motion data. Using such database instead of direct EMG measurement is justified by the fact that muscle tension distribution is relatively highly correlated even among different subjects. For each new motion frame, we search for a similar entry in the database and use the corresponding muscle tension distribution to estimate the current muscle tensions. We demonstrate that the muscle tensions obtained by this approach is much closer to the result using the EMG data than that using pure numerical optimization, even when the database is constructed from other person's data.

In previous work, the authors have been developing a stochastic model based approach for on-line segmentation of whole body human motion patterns during human motion observation and learning, using a simplified kinematic model of the human body. In this paper, we extend the proposed approach to larger, more realistic kinematic models, which can better represent a larger variety of human motions. These larger models may include spherical in addition to revolute joints. We examine the effects on segmentation performance due to motion representation choice, and compare the segmentation efficacy when Cartesian or joint angle data is used. The approach is tested on whole body human motion data modeled with a 42DoF kinematic model. The results indicate that Cartesian data seems to correspond most closely to the human evaluation of segment points. The experiments also demonstrate the efficacy of the segmentation approach for large kinematic models and a variety of human motions.

Since humanoid robots have similar body structures to humans, a humanoid robot is expected to perform various dynamic tasks including object manipulation. This research focuses on issues related to learning and performing object manipulation. Basic motion primitives for tasks are learned from observation of human's behaviors. An object manipulation task is divided into two types of motion primitives, which are represented as hidden Markov models (HMMs): one for a body motion primitive and the other for the relation between the object and body parts, which manipulate the object. When performing a task, a natural whole body motion is associated from an object motion by using learned motion primitives. Furthermore, the associated body motion is reshaped in both spatial and temporal space, in a more precise way. The reshaping in spatial space is realized in two stages by a feedback control policy learned with reinforcement learning and by constrained inverse kinematics. Key features like end-effectors for manipulation and timing for a task are extracted and used for the feedback control policy learning. The reshaping in temporal space is realized by comparing a predicted and observed object motion speed.

In this paper, a position based impedance controller (i.e. admittance controller) is designed by utilizing measurements of a force/torque sensor, which is mounted at the robot's base. In contrast to conventional force/torque sensing at the end-effector, placing the sensor at the base allows to implement a compliant behavior of the robot not only with respect to forces acting on the end-effector but also with respect to forces acting on the robot's structure. The resulting control problem is first analyzed in detail for the simplified one-degree-of-freedom case in terms of stability and passivity. Then, an extension to the Cartesian admittance control of a robot manipulator is discussed. Furthermore, it is shown how the steady state properties of the underlying position controller can be taken into account in the design of the outer admittance controller. Finally, a simulation study of the Cartesian admittance controller applied to a three-degrees-of-freedom manipulator is presented.

This paper describes an novel approach towards linguistic processing for robots through integration of a motion language module and a natural language module. The motion language module represents association between symbolized motion patterns and words. The natural language module models sentences. The motion language module and the natural language module are graphically integrated. The integration allows robots not only to interpret observed motion as a sentence but also to generate motion with a sentence. This paper proposes incremental learning algorithm of association between symbolized motion patterns and words. The incremental learning is required for robot to autonomously develop the linguistic skill. The algorithm can be derived from optimization of the motion language module under stochastic constraints such that the associative probability of a new training pair composed of symbolized motion pattern and sentence becomes larger. Test of interpreting observed motion as sentences demonstrates the validity of the proposed incremental learning algorithm.

Mass parameters of the human body segments are mandatory when studying motion dynamics. In orthopedics, biomechanics and rehabilitation they are of crucial importance. Inaccuracies their value generate errors in the motion analysis, misleading the interpretation of results. No systematic method to estimate them has been proposed so far. Rather, parameters are scaled from generic tables or estimated with methods inappropriate for in-patient care. Based on our previous works, we propose a real-time software and its interface that allow to estimate the whole-body segment parameters, and to visualize the progresses of the completion of the identification. The visualization is used as a visual feedback to optimize the excitation and thus the identification results. The method is experimentally tested and obtained results are discussed.

This paper proposes an on-line, interactive approach for incremental learning and visualization of full body motion primitives from observation of human motion. The human demonstrator motion is captured in a motion capture studio. The continuous observation sequence is first partitioned into motion segments, using stochastic segmentation. Motion segments are next incrementally clustered and organized into a hierarchical tree structure representing the known motion primitives. At the same time, the sequential relationship between motion primitives is learned, to enable the generation of coherent sequences of motion primitives. An on-line visualization system is also developed to allow the demonstrator to visualize the motion database and the motion primitives learned by the system, thus giving the demonstrator insight into the learning process and the ability to interactively modify the demonstration based on the current state of the knowledge base. The developed system has many potential applications for motion analysis, prediction and imitation learning for humanoid robots.

This paper describes a novel approach for incremental learning of human motion pattern primitives through online observation of human motion. The observed time series data stream is first stochastically segmented into potential motion primitive segments, based on the assumption that data belonging to the same motion primitive will have the same underlying distribution. The motion segments are then abstracted into a stochastic model representation and automatically clustered and organized. As new motion patterns are observed, they are incrementally grouped together into a tree structure, based on their relative distance in the model space. The tree leaves, which represent the most specialized learned motion primitives, are then passed back to the segmentation algorithm so that as the number of known motion primitives increases, the accuracy of the segmentation can also be improved. The combined algorithm is tested on a sequence of continuous human motion data that are obtained through motion capture, and demonstrates the performance of the proposed approach.

A completely stepwise online pedipulation planning method is proposed. It is an analytical approach based on the general solution of the equation of motion of an approximate dynamical biped model whose mass is concentrated at the center of mass. A physically feasible referential trajectory with a constraint about the reaction force taken into account is planned only in one interval by relaxing the boundary condition, namely, by admitting a certain level of error between the desired and actually reached states, and discontinuity of zero-moment point at each end of the interval. It potentially creates responsive motions that require strong instantaneous acceleration. A semiautomatic continual pedipulation planning method is also presented. It generates a referential trajectory of the whole body only from the next desired foot placement. The validity of the proposed method is ensured through experiments with a small anthropomorphic robot.

This paper describes a novel approach for incremental learning of human motion pattern primitives through on-line observation of human motion. The observed motion time series data stream is first stochastically segmented into potential motion primitive segments, based on the assumption that data belonging to the same motion primitive will have the same underlying distribution. The motion segments are then abstracted into a stochastic model representation, and automatically clustered and organized. As new motion patterns are observed, they are incrementally grouped together based on their relative distance in the model space. The resulting representation of the knowledge domain is a tree structure, with specialized motions at the tree leaves, and generalized motions closer to the root. The tree leaves, which represent the most specialized learned motion primitives, are then passed back to the segmentation algorithm, so that as the number of known motion primitives increases, the accuracy of the segmentation can also be improved. The combined algorithm is tested on a sequence of continuous human motion data obtained through motion capture, and demonstrates the performance of the proposed approach.

This paper proposes a method to recover missing data during observation by factorial hidden Markov models (FHMMs). The fundamental idea of the proposed method originates from the mimesis model, inspired by the mirror neuron system. By combining the motion recognition from partial observation algorithm and the proto-symbol based duplication of observed motion algorithm, whole body motion imitation from partial observation can be achieved. The algorithm for missing data recovery uses the same basic strategy as the whole body motion imitation from partial observation, but requires more accurate spatial representability. FHMMs allow for more efficient representation of a continuous data sequence by distributed state representation compared to hidden Markov models (HMMs). The proposed algorithm is tested with human motion data and the experimental results show improved representability compared to the conventional HMMs.

In this paper a compliance control law for kinematically redundant manipulators is proposed. The controller contains a Cartesian compliance part and a nullspace compliance part which are complemented by a power-conserving decoupling term. The approach deliberately avoids inertia shaping in order to obtain a control law which does not require the measurement of external forces and becomes less sensitive with respect to model uncertainties. While the controller formulation explicitly uses nullspace velocity coordinates, no integration of these velocities is required. Except for the kinematic singularities of the manipulator’s Jacobian matrix, no further algorithmic singularities are introduced. Asymptotic stability of the closed-loop system is shown by utilizing semi-definite Lyapunov functions. Finally, a short planar simulation study is presented which validates the effectiveness of the approach.

By controlling complex robotic systems one often has to cope with the situation that different sub-systems are interfaced and controlled by different computers. In this paper the problem of coordinated control of such a system with distributed control structure is addressed. In particular one must handle the transmission delays in the communication between the different computers, which can be considered small but not negligible, since also small delays in the transmission of power variables violate the passivity and therefore may lead to instability. In this paper the wave variables concept is applied to handle the delays and is used in combination with a virtual inertia for designing a Cartesian compliance controller. Therefore, in particular the steady state properties of the wave variable based communication is of interest and leads for the case of small delays to the analogy with a flexible joint robot. In a second step the virtual inertia is eliminated in order to approximate the desired closed loop behavior better. Finally, some simple planar simulations are presented which validate the proposed approach.

When simulating and controlling robot dynamics it is necessary to know the inertial parameters and the joint dynamics accurately. As these parameters are usually not provided by manufacturers, identification is then an essential step in robotics. In addition with the up coming wide-spreading of humanoid robots in the society the identification of humanoid dynamics has became mandatory to insure safety. This paper proposes a method to estimate humanoid robots inertial parameters using a minimal set of sensors. Only joint angles and external forces information are required. Simulations have provided exciting trajectories that are reproduced on a small-size humanoid robot. Experimental results are given.

This paper develops an approach for on-line segmentation of whole body human motion patterns during human motion observation and learning. A Hidden Markov Model is used to represent the incoming data sequence, where each model state represents the probability density estimate over a window of the data. Based on the assumption that data belonging to the same motion primitive will have the same underlying distribution, the segmentation is implemented by finding the optimum state sequence over the developed model. The basic algorithm is modified to add the capability for modifying the model based on known motion primitives. The inclusion of such scaffolding motion primitives can improve the performance of the basic segmentation algorithm. The modified algorithm is tested on a corpus of continuous human motion data to show the efficacy of the proposed approach.

Since humanoid robots have similar body structures to humans, they are expected to perform various tasks including tool-use manipulation tasks instead of humans. This research studies on learning and performing tool-use manipulation tasks. For tool-use manipulations, understanding the relation between tool motion and whole body motion is crucial. In this paper, a tool-use motion model is designed with tool knowledge and body motion knowledge. The authors propose a method which enables a humanoid robot to associate whole body motion from tool knowledge by adopting the mimesis method from partial observations [1]. When a specific tool trajectory of a tool-use motion is given, appropriate hand motion is associated. From the calculated hand motion, appropriate whole body motion is associated successively. The proposed algorithm is implemented on a humanoid robot.

This paper presents our microscope image stabilization system for in vivo microscopy. This work is a novel robotic application. In vivo microscopy is very in demand due to its potential impact on biological research [1]. However, it has turned out that in vivo microscopy is significantly disturbed by the motion of the imaged tissue of a living animal. The proposed system virtually removes the unwanted motion by synchronizing the motion of the objective lens with it. In order to realize this idea, we have developed a simple contact-type sensor for estimating the motion of organs, and also made a motion compensator for moving the objective lens. Sensing and compensating have been accomplished for 3-D translational motion. Not only laboratory tests with artificial motion, but also in vivo experiments show the successful motion canceling effect of the proposed system.

This paper describes an imitative learning of driving time series data for intellectual cognition toward future automobiles. The driving pattern primitives consisting of states of the environment, vehicle and driver are symbolized by hidden Markov models (HMMs), which can be used for both recognition and generation of the driving patterns. The relationship among the HMMs can be represented by locating the HMMs in a multidimensional space. The contribution of each variable to the HMM space can be analyzed such that important variables can be selected out of the driving data in order to reduce the size of the HMMs. Moreover, this paper presents a hierarchical model with the HMMs abstracting the primitive driving patterns in the lower layer, and another HMMs abstracting the longterm contextual driving patterns which are representation in the HMM space. Tests with a driving simulator and a actual vehicle demonstrate not only the validity of symbolization of driving pattern primitives, recognition and generation, but also availability of key feature selection. The extended hierarchical model is also proved to have a potential to predict the driving data appropriately.

This paper presents a dynamics simulator that can handle complex robotic systems including position-controlled joints and closed kinematic chains. We first extend our prevous algorithm for linear-time forward dynamics algorithm to handle closed kinematic chains. The extended algorithm is formally presented for the first time. We then present another extension that allows position-controlled joints, whose angles exactly follow the reference by perfect servo controllers. This feature is often useful for simple trial simulations only using joint angle commands because the user does not have to design low-level servo controllers for simulation. The simulation can also be performed without precise friction parameters. The algorithm is tested on a humanoid robot having toe joints with four-bar linkage structure.

This paper presents image stabilization for microscopy using horizontal visual feedback control of the objective lens through a five-bar linkage and piezoelectric actuators, and its application to in vivo imaging. Even very small in vivo motion due to heartbeat and breathing makes microscopic observation difficult by blurring the microscope image or impossible by sending a region of interest out of view. In order to remove those unwanted effects of the motion, we have introduced motion-canceling robotic technologies into microscopy. Our image stabilization system through motion-canceling provides users with stabilized image sequences with respect to trembling of in vivo subjects. The developed image stabilization system, in term of robotics, corresponds to a visual feedback control system that consists of a robotic mechanism and a high-speed vision. A high-speed camera installed in the microscope detects the motion of the in vivo subject having topically applied fiducials. To virtually cancel this motion, we move the objective lens, synchronizing the motions of the subject and the lens to remove the relative motion between the two. As a result, we observe motion-free images to m. This technology is one of the very demanding technologies in biological research for in vivo observation with high resolution. In this paper, we verify the effectiveness of the developed system through in vivo experiments.

Optical motion capturing systems are widely used to acquire human beings' motion patterns in humanoid imitation learning research. However, optical motion capturing systems have a restricted movable area. This paper proposes the HMM based mimesis scheme using a monocular camera mounted on a humanoid. This scheme releases the restriction of movable area and enables imitation in daily life environments. Also, natural human-robot-interaction is expected during imitation. From two-dimensional image sequences of the demonstrator's motion, the demonstrator's pose and motion is estimated and recognized through the mimesis model and the humanoid generates its joint motor commands for imitation in 3D space. The feasibility of the proposed scheme is demonstrated by simulation.

Motion capture systems are used to obtain motion data such that humanoid robots or computer graphics (CG) characters can behave naturally. However, it has proven to be hard not only to modify the capture data without losing its reality but also to search for the required capture data in a lot of capture data. In this paper, we provide a solution to these problems based on our previous work on symbolization of motion patterns for developing humanoid intelligence. Similar motion sequences in the database are abstracted as a symbol, which will be applied to searching motion patterns in the database similar to a given motion. This paper also introduces a method for building a stochastic symbol-word mapping model utilizing the word labels provided by the operator during motion capture sessions. This model converts a input sequence of words into a sequence of symbols, and then allows the capture database both to be searched for capture data corresponding to the input (a sequence of words) and to provide the users with new motion data generated by the symbols. Finally, we apply analogy of symbols to establishing the database in order to provide an appropriate motion data in response to an unsupervised sequence of words and then demonstrate the validity of analogy theory.

This paper describes an improved methodology for human motion recognition and imitation based on Factorial Hidden Markov Models (FHMM). Unlike conventional Hidden Markov Models (HMMs), FHMMs use a distributed state representation, which allows for more efficient representation of each time sequence. Once the FHMMs are trained with exemplar motion data, they can be used to generate sample trajectories for motion production, and produce significantly more accurate trajectories compared to single Hidden Markov chain models. Due to the additional information encoded in FHMMs models, FHMM models have a higher Kullback- Leibler distance compared to single Markov chain models, making it easier to distinguish between similar models. The efficacy of using FHMMs is tested on a database of human motions obtained through motion capture. The results show that FHMMs provide better generalization to new data when compared to conventional HMMs during motion recognition, as well as providing a better fit for generated data.

This paper proposes a 3D motion recovery method from monocular images by statistical inference. The fundamental idea of the paper originates from the mimesis model, inspired by the mirror neuron system. The mimesis model is extended to include motion understanding from monocular image sequences and to imitate whole-body motion patterns in 3D space. In order to achieve this goal, (1) conversion of 3D motion database, represented in probabilistic form, into various spaces is adopted. (2) A vector field approach is developed for natural motion understanding. (3) With the particle filter, a demonstrator’s pose is estimated.

Boundary condition relaxation (BCR) method proposed by the authors (2005) is enhanced to enable 3D hopping motion plannings from arbitrary initial conditions. The original BCR has an advantage that stepwise legged motion planning is realized in online by accepting an error from the desired goal state of the center of mass (COM) and discontinuity of the trajectory of the zero-moment point (ZMP). The main difficulty of the enhancement lies on that heterogeneous piecewise equations of motions have to be handled at once, and seamless conditioning about the angular-momentum conservation and the body attitude before/after contact phase changing have to be achieved. For the former issue, multiple boundary conditions of piecewise differential equations are set up and solved in the same way with the original BCR. They are based on an approximately mass-concentrated biped model, so that contact state transition and severe time constraints are dealt with at low computational cost. Vertical-horizontal interference of COM trajectory is also taken into account by applying numerical solution of the initial value problem of differential equations. For the latter, a Jacobian-based inverse kinematics with a continuously-varying weight-blending in accordance with the shift of the contact state is presented.

hierarchical model incorporating motion patterns, proto symbols and words is proposed. The proto symbols abstract motion patterns, while the words are associated with the proto symbols stochastically. This paper describes the construction of a word space, where words are located in a multidimensional space based on dissimilarities among the words. The dissimilarity between two words can be calculated by using association probabilities that the words generate motion proto symbols. The word space encapsulates relations among the words such as similar or dissimilar pairs of words. The word space also allows motion recognition based on words. The validity of the constructed word space is demonstrated on a motion capture database. Moreover, the addition of the word associations is found to change the conventional proto symbol space so that the discrimination among the proto symbols is improved.

We present a solution to estimate in-vivo the joint dynamics of the human limbs during passive movements. The method is based on well-known modelling and approach used in Robotics that allow simultaneous multi-joint estimation. The modelling of the human body and the human joint as well as the method are described. The experimental set-up based on the use of an optical motion capture system is detailed. Three types of movements are recorded and used to perform the identification. We concluded that designed movements and movements from clinical diagnosis of neuromuscular diseases are good to perform the identification; however swing of the arms during normal walk does not provide enough excitation to obtain consistent results.

This paper describes a novel algorithm for autonomous and incremental learning of motion pattern primitives by observation of human motion. Human motion patterns are abstracted into a Hidden Markov Model representation, which can be used for both subsequent motion recognition and generation, analogous to the mirror neuron hypothesis in primates. As new motion patterns are observed, they are incrementally grouped together using hierarchical agglomerative clustering based on their relative distance in the HMM space. The clustering algorithm forms a tree structure, with specialized motions at the tree leaves, and generalized motions closer to the root. The generated tree structure will depend on the type of training data provided, so that the most specialized motions will be those for which the most training has been received. Tests with motion capture data for a variety of motion primitives demonstrate the efficacy of the algorithm.

This paper proposes a computational method of contact forces working between multibody system and environment in forward dynamics based on both the microbody-deformation model and macro contact model. The combination of them simultaneously prevents the simulation from excess penetration in micro contact model and chattering in macro contact model. The difficulty lies on how to absorb the difference of duration between them. This problem is solved through the introduction of a timestep-dependent variable damper for a micro contact model and of error-norm minimization for a macro contact model

Communication skill is essential for social robots in various environments such as homes, offices, and hospitals, where the robots are expected to interact with humans. In this paper, we model the primitive nonverbal communication between two persons by mimetic communication model. The model consists of three groups of hidden Markov models (HMMs) hierarchically combined to recognize motions of the human and to generate the interactive motions of the robot. HMMs in the lower layer abstract the motion patterns and HMMs in the upper layer represent the interaction patterns. We demonstrate the validity of this model through kick boxing match between a motion-captured human and humanoid robot, where the robot can autonomously generate its motion in response to attacks by the human

Human upper limb joints dynamics is very important in the fields of humanoid robotics, medical robotics as well as medical research. To make human-like passive movements of the arms when walking humanoid robot arms must have similar dynamics to the human arms, even more if this arm is to be used as a prosthesis. Moreover medical diagnosis of muscle or neuro-motor diseases are based on a visual qualitative estimation of joint passive stiffness. There is a pressing need in human body dynamics characterization and especially in subject specific characterization. In this paper a solution to estimate in-vivo the passive dynamic of the arm joint is proposed. It is based on the use of the musculo-skeletal description of the human body and its kinematics computation. The linear passive joint dynamics: stiffness, viscosity and friction, is then estimated with least squares method. Acquisition of movements both designed for estimation or from medical diagnosis check-up, are achieved with motion capture studio only (no pain, no distress on subject). Experimental results for three valid subject are given

This paper presents a stable penalty-based model for simulating frictional contacts between many complex objects. The major advantage of our model is that it solves the problems in implementing Coulomb's friction model for computer simulation: iterative computation and slip velocity threshold. We also introduce a robust method for computing the normal vector and penetration depth at each contact point of a pair of interpenetrating polygonal objects. We demonstrate the validity and usability of the model by comparing the simulation results with closed-form solutions of Coulomb's friction model and conventional friction model, as well as performing dynamics simulation of highly complex scenes with tens of objects composed of thousands of polygons

Generally, imitation of a motion means generation of a close motion to the observation. Moreover, it means that conversion into its own motion, which is adoptable to its body structure, by integrating with its prior knowledge. From this perspective, a new imitation scheme is proposed. The scheme is based on hidden Markov models by employing Viterbi algorithm. The proposed scheme enables to imitate a new observed motion without learning the motion by applying its prior knowledge. Online motion primitive acquisition method is considered. Evaluation factors, such as inheritance coordinate and matching error, are introduced to evaluate imitation performance. The feasibility of the proposed scheme is demonstrated by simulation on a 20 degrees of freedom humanoid robot configuration with the evaluation factors

The human motions are generated through the interaction between the body and its environments. The information processing system defines the current motion using the signal feedback of the body state and environments. The motion pattern dose not exits a priori but emerges as the result of the entrainment phenomenon for the dynamics of the information processing, the human body and its environments. In this paper, based on the dynamics-based information processing system, we propose the motion emergency system design method for a humanoid robot designing a dynamical system that has an attractor considering the robot body dynamics. From the control engineering point of view, the proposed method designs a controller that stabilizes the robot to an equilibrium trajectory.

A fast online gait planning method is proposed. Based on an approximate dynamical biped model whose mass is concentrated to COG, general solution of the equation of motion is analytically obtained. Dynamical constraint on the external reaction force due to the underactuation is resolved by boundary condition relaxation, namely, by admitting some error between the desired and actually reached state. It potentially creates responsive motion which requires strong instantaneous acceleration by accepting discontinuity of ZMP trajectory, which is designed as an exponential function. A semi-automatic continuous gait planning is also presented. It generates physically feasible referential trajectory of the whole-body only from the next desired foot placement. The validity of proposed is ensured through both simulations and experiments with a small anthropomorphic robot.

This paper presents a method for capturing detailed human motion by using a suit covered with retroreflective mesh. We can attach huge number of markers on the subject without replacing the hardware of current passive optical motion capture systems. Compared to normal motion capture using spherical markers, the connectivity information of the mesh can be used to improve the efficiency and accuracy of the reconstruction process. As a result, the system can achieve faster and more precise measurement of hundreds of markers on the subject than other approaches such as using natural image or 3D scanning. The total computation time required to reconstruct the 3D mesh information including 408 markers (intersections) is 65.5 ms, allowing realtime motion capture at 15 fps.

The goal of this research is to enable precise estimation of human muscle forces in whole-body motions based not only on physiological muscle model but also on the equation of motion. The potential application areas include human-machine interface, medicine, biomechanics, and computer animation. Towards this goal, in this paper we discuss the inverse dynamics of musculoskeletal human model using the data from electromyogram (EMG) and force sensor. The inverse dynamics of musculoskeletal human models is formulated as an optimization problem subject to equality and inequality conditions taking into account of the equation of motion and muscle model from physiology literature. We evaluate the developed algorithm on a complex musculoskeletal model with 366 muscles driving a skeleton with 155 degrees of freedom.

In the age of PR (personal robot), the demand for various and complicated exterior designs has become greater. This paper focuses on improving the designability of the exterior of robots. A main cause of discouraging the designability is that the exterior and mechanism designs are not independent. We try to lower the mutuality between them by proposing a fabricating method of humanoid covers with insert molding. This method prevents weight excess and shrinkage of the movable ranges caused by attaching the cover and allows complicated shape of the cover, but has three major problems to be solved. Firstly, the restoring joint torque caused by the deformation of the cover should be estimated. Secondly, the cables should be arranged inside the cover so as not to be tensioned too much. Thirdly, an appropriate way of heat radiation is needed. In this paper, design methods of a humanoid with insert-molded cover such as the equations to estimate the restoring torque, the cable arrangement and the way of heat radiation are described. Finally, the prototype now being designed is mentioned.

Bifurcations of attractors take place in associative neural networks with nonmonotonic activation functions, depending on the degree of correlations between stored patterns and the parameter of nonmonotonicity. We describe the bifurcations when auto-correlation based feature vectors of motion patterns of humanoid robots, which are hierarchically correlated, are stored. Also, we describe a memory system which utilizes the neural network dynamics and hierarchically maintains specific and conceptual memories of motions of humanoid robots. The level of abstraction is controlled by a parameter in the retrieval phase without changing the connection weights.

In this paper, a new mimesis scheme is proposed. This scheme enables for a humanoid to imitate human's motion even though the humanoid cannot see human's whole-body motion and the humanoid has not seen the exactly same motion so far. Mimesis framework is based on continuous hidden Markov model. Viterbi algorithm is applied in order to generate more various motion patterns than the number of existing hidden Markov models. In order to imitate other's motion in a smooth way, a smoothing technique in generation problem is realized. The feasibility of this method is demonstrated by simulation on 20 degrees of freedom humanoid robot configuration.

The motivation of this research is to compute internal perspective of humans through external observation. In this paper, we propose method for analyzing neural information through motion measurement. The method is on the bases of the anatomy and physiology of somatic and spinal nervous system. Muscles are classified by the innervated nerves originate from the spinal cord. The somatotopic organization inside the ventral horn of the spinal cord is utilized for topological structure of the spinal neural information. Time series of images which represent distribution of somatic information inside the spinal cord were successfully obtained through measurement and computation for sword swinging 'kesagiri' motion. The coordination of the motion at spinal level was analyzed. The proposed method provides fundamental for motor learning support system.

A design methodology to build miniature humanoid robots is discussed. Although light and small bodies would make aggressive types of motion experiments much safer and smoother, they would often cause self-collisions and even restrict the space to mount mechatronic components. In order to defeat some kinematic difficulties including the former issue, a technique to modularize and assign joints is proposed through our prototyped robot. And, as a solution against the latter issue, a portable core control unit which stores a stand-alone electronic system is also introduced through the second version of our humanoid, whose system centers around it.

Since a biped humanoid inherently suffers from instability and always risks tipping itself over, ensuring high stability and reliability of walk is one of the most important goals. This paper proposes a walk control consisting of a feedforward dynamic pattern and a feedback sensory reflex. The dynamic pattern is a rhythmic and periodic motion, which satisfies the constraints of dynamic stability and ground conditions, and is generated assuming that the models of the humanoid and the environment are known. The sensory reflex is a simple, but rapid motion programmed in respect to sensory information. The sensory reflex we propose in this paper consists of the zero moment point reflex, the landing-phase reflex, and the body-posture reflex. With the dynamic pattern and the sensory reflex, it is possible for the humanoid to walk rhythmically and to adapt itself to the environmental uncertainties. The effectiveness of our proposed method was confirmed by dynamic simulation and walk experiments on an actual 26-degree-of-freedom humanoid.

In this paper, we develop a 3-degree of freedom (DOF) mechanism for humanoid robots, which we call the cybernetic shoulder. This mechanism imitates the motion of the human shoulder and does not have a fixed center of rotation, which enables unique human-like motion in contrast to the conventional design of anthropomorphic 7-DOF manipulators that have base three joint axes intersecting at a fixed point. Taking advantage of the cybernetic shoulder's closed kinematic chain, we can easily introduce the passive compliance adopting the elastic members. This is important for the integrated safety of humanoid robots that are inherently required to physically interact with the human.

This paper discusses the design issue of general closed kinematic chains focusing on analysis of constraints and elastic deformations. Closed kinematic chains are considered more advantageous in rigidity, power-output, and accuracy than open kinematic chains. However, it is not much stressed that closed kinematic chains are so sensitive to machining errors that a tenth of a millimeter of error might result in jamming and immobility. To avoid this critical problem, closed kinematic chains tend to be designed with play that reduces their native advantages. If 3D-CAD systems are equipped with a mathematical tool that evaluates machining accuracy, elasticity, and constraints, it will significantly assist the skill of designers and extend the field of applications of closed kinematic chains. In this paper, we clarify machining errors that are absorbable as errors in unactuated joints. These kind of errors are permissible. Mobility analysis in the presence of unabsorbable machining errors is discussed taking account of elastic deformations and strain energy. The mechanism can move smoothly even with machining errors if the strain energy remains less fluctuate along its motion locus. The established mathematical method of mobility analysis is applied to the design of a closed kinematic chain and used in practice to fabricate a medical robot system.

In this paper, we design a continuous symbol space using a dynamics-based information processing system. One point in the symbol space decides the vector field in the motion space that generates the cyclic motion and the continuous motion transition of the robots. Because the motion of the state vector in the symbol space is defined by a dynamical system, the spatial and temporal continuous information processing system is realized.

This paper proposes the mathematical model of deformation for gel robots and develops the pattern formation theory. The robots are made of surfactant-driven ionic polymer gel in constant electric fields, which is a typical electroactive polymer gel containing poly 2-acrylamido-2-methylpropane sulfonic acid (PAMPS). A beam of gel in uniform electric fields develops wave forms through penetration of the surfactant solution. The model is to be built on the hypothesis of adsorption-induced deformation. The mechanism of wave-shape pattern formation is then analyzed utilizing the model. The results of this study provide the foundation to develop deformable machines with virtually infinite degrees of freedom.

This paper describes an efficient algorithm to generate a set of local optimal paths between two given end points in cluttered environments or over open terrain. The local optimal paths are selected from the set of shortest constrained paths through every node (one for each path) in the graph, generated by running twice a "single-source" search. The initial set of the shortest constrained paths spans the entire search space and includes local optimal paths with costs equal or better than the longest constrained path in the set. The search for the optimal path is transformed to a search for the best path in each homotopy class generated by this search. The initial search is of complexity O(nlogn), and the pruning procedure is O(nmlogm), where n is the number of nodes and m is the number of homotopy classes generated by this search. The algorithm is demonstrated for motion planning on rough terrain.

Today optical motion capture system is becoming an essential tool for motion analysis, synthesis, and character animation. This paper focuses on improving the ability of current passive optical motion capture systems, which provides the highest precision and flexibility with the lowest interference among current motion capture technologies but has two major drawbacks. Firstly, it usually requires expensive post-processing computation including reconstruction and labeling. The second problem is that it is difficult to achieve both high precision and high frame rate at the same time due to the limitation of data transmission rate, that is, high-resolution cameras have low frame rate and vice versa. In this paper, we try to solve these problems by combining cameras of different types that complement the limitations of each others. The marker positions measured by the high-resolution cameras correct the low-precision data from high-speed cameras, which in turn helps real-time tracking of markers by inserting new data at higher frame rate. The proposed method is implemented on a PC cluster and experimental results show that we can obtain high-precision data even for high-speed motions. We also demonstrate the real-time joint angle computation using the real-time tracking capability.

This paper describes a new search algorithm, the Dual Dijkstra Search. From a given initial and final configuration, Dual Dijkstra Search finds various paths which have different topologies simultaneously. This algorithm allows you to enumerate not only the optimal one but variety of meaningful candidates among local minimum paths. It is based on the algorithm of Dijkstra, which is popularly used to find an optimal solution. The method consists of two procedures: First computes local minima and ranks the paths in order of optimality. Then classify them with their topological properties and take out only the optimal paths in each groups. Computed examples include generating collision-free motion along 2D space and motion planning of 3-DOF robot. We also proposed the idea of motion compression, which simplifies the high dimensional motion planning problem. Together with this idea, we applied Dual Dijkstra Search to 7-DOF arm manipulation problem and succeeded in obtaining variety of motion candidates.

In this paper, we develop two mechanisms for improving humanoid robot motions. The double spherical joint is a six DOF mechanism whose axes intersect in one point. This mechanism is used for humanoid hip joints with a waist joint function without increasing the actuator. The backlash clutch realizes a high torque driving and a free joint using backlash mechanisms is used for knee joints. The free mode will play a role in humanoid behavior that is dynamically coupled with the environment. The humanoid robot with these two mechanisms is developed and results of preliminary experiments are shown.

Being expected as the utilities in the future, humanoid robots should be given much higher mobility. A seamless transition between contact and aerial phase is essential to behave robustly against disturbance in the real environment, and to expand the range of their activities and perform a variety of motion. Manipulation of both the contact condition and the external force is the key issue to enhance the mobility of humanoids since they are driven by the external force converted from the inner force through the interaction with the environment. The difficulty lies on the complexity of their dynamics so that they consist of number of degrees of freedom and their structures vary in accordance with contact phase transition. We propose variable impedant inverted pendulum (VIIP) model control, which allows one to handle the external force rather easily. The advantage of the proposed is that it is invariant on contact phase so that both cases in contact and in aerial are treated in the unified way. It also reduces the amount of computation. Thus, quick responsive motion of the robot can be practically achieved. We verified the effect of the controller in computer simulation, using a small humanoid robot model.

Since a humanoid robot takes the morphology of human, users as pilots will intuitively expect that they can freely manipulate the humanoid extremities. However, it is difficult to simultaneously issue such multiple control inputs to the whole body with simple devices. It is useful for motion pattern generation to get mapping functions bidirectionally between a large number of control inputs for a humanoid robot and a small number of control inputs that a user can intentionally operate. For the purpose of generation of voluntary movement of humanoid extremities, we introduce hierarchical NLPCA neural networks that forms low dimensional variables out of multi-variate inputs of joint angles. The problem is to find common space that affords unified manipulable variables not only for specific motion like walk but also multiple whole body motion patterns. The interesting result is shown that 1 dimensional inputs can generate an approximate walking pattern, and also 3 dimensional inputs does 9 types of motion patterns.

Memory of motion patterns as data, comparison of a new motion pattern with data, and playback of one from the data are inevitably involved in the information processing of intelligent robot systems. Such computation forms the computational foundation of learning, acquisition, recognition, and generation process of intelligent robotic systems. In this paper, we propose to apply the continuous hidden Markov model to establish the computational foundation, using which one obtains the specified number of keyframes and their probability distributions. The keyframes are optimally selected to maximize the likelihood. The probability distributions are to be used to compute comparison and playback. The proposed method is applied to the motion data of a humanoid robot as well as the time series image data, and its validity is to be discussed.

In this paper, we develop the on-line design method and the hierarchical design method of dynamics based information processing system for the robot intelligence. By using the forgetting parameter, dynamics memorizes a new robot motion forgetting an old motion, which means the plasticity of the system. The hierarchical structure enables information processing for complex and continuous environment. We implement the proposed method to a humanoid robot and realize the motion generation and transition.

In this paper, we describe the concept and implementation of a dynamics filter, an online, full-body motion generator that converts a physically infeasible reference motion into a feasible one for the given human figure. Our implementation of the dynamics filter only uses time-local information, that is, does not require the whole motion sequence in advance. Therefore, the reference motion may be changed online in response to the interaction with a human or the environment. The dynamics filter is implemented based on an efficient rigid-body collision/contact model. This model itself provides an efficient algorithm for dynamics simulation of collisions and contacts. We demonstrate the power of the dynamics filter by several example motions that use motion capture data as a reference.

This paper presents a computational technique for creating whole-body motions of human and animal characters without reference motion. Our work enables animators to generate a natural motion by dragging a link to an arbitrary position with any number of links pinned in the global frame, as well as other constraints such as desired joint angles and joint motion ranges. The method leads to an intuitive pin-and-drag interface where the user can generate whole-body motions by simply switching on or off or strengthening or weakening the constraints. This work is based on a new interactive inverse kinematics technique that allows more flexible attachment of pins and various types of constraints. Editing or retargeting captured motion requires only a small modification to the original method, although it can also create natural motions from scratch. We demonstrate the usefulness and advantage of our method with a number of example motion clips.

It is needed for humanoid to acquire not only just a trajectory but also aim of the behavior and symbolic information during behavior development. We (2001) have proposed the mimesis system as a framework of synchronous learning model for behavior acquisition and symbol emergence. However, the motion elements which are fundamental representation of behavior have stood on the unsuitable assumption that they are given without taking the robot embodiment and dynamics into consideration. In this paper, the design theory of motion elements with consideration of the embodiment are shown, and novel methods of realization of the mimesis for real humanoids is proposed.

This paper introduces an open architecture humanoid robotics platform (OpenHRP) on which various building blocks of humanoid robotics can be investigated. OpenHRP is a virtual humanoid robot platform with a compatible humanoid robot, and consists of a simulator of humanoid robots and motion control library for them which can also be applied to a compatible humanoid robot as it is. OpenHRP is expected to initiate the exploration of humanoid robotics on an open architecture software and hardware, due to the unification of the controllers and the examined consistency between the simulator and a real humanoid robot.

This paper presents the real time processing of optical motion capture with pan-tilt camera tracking. Pan-tilt camera tracking expands the range of capturing field dynamically. The asymmetrical marker distribution and polyhedra search algorithm realize robust labeling against missing markers. The algorithm is developed for parallel cluster computation and enables real time data processing. Experimental results demonstrate the effectiveness of the system.

Use of compliance in muscle is the inherent skill of a human. By using the potential energy charged in the compliant members, we can skillfully equalize the characteristics of muscles and body. Integrating the skill of compliance will provide robots with higher mobility, dexterity and safety and extends the fields of applications. The main research issues of the skill of compliance are tuning passive compliance, planning compliant motion and designing control law. To achieve the skill, we focus on the planning compliant motion considering the kinetic energy. In this paper, we propose to design a compliant motion through iterative model identification and motion design. A humanoid robot with passive compliance is used to integrate the skill of compliance and shows fast swing charging and discharging the kinetic energy.

For the development of an intelligent robot with many degrees-of-freedom, the reduction of the whole body motion and the implementation of the brain-like information system is necessary. We propose a reduction method of the whole body motion based on singular value decomposition and a design method of the brain-like information processing system using a nonlinear dynamics network with polynomial configuration. By using the proposed method, we design the humanoid whole body motion that is caused by the input sensor signals.

A humanoid robot is expected to be a rational form of machine to act in the real human environment and support people through interaction with them. Current humanoid robots, however, lack in adaptability, agility, or high-mobility enough to meet the expectations. In order to enhance high-mobility, the humanoid motion should be generated in real-time in accordance with the dynamics, which commonly requires a large amount of computation and has not been implemented so far. We have developed a real-time motion generation method that controls the center of gravity (COG) by indirect manipulation of the zero moment point (ZMP). The real-time response of the method provides humanoid robots with high-mobility. In the paper, the algorithm is presented. It consists of four parts, namely, the referential ZMP planning, the ZMP manipulation, the COG velocity decomposition to joint angles, and local control of joint angles. An advantage of the algorithm lies in its applicability to humanoids with a lot of degrees of freedom. The effectiveness of the proposed method is verified by computer simulations.

An efficient parallel algorithm for forward dynamics computation of human figures is proposed. The algorithm is capable of handling any kinematic chains including structure-varying ones. The asymptotic complexity of the algorithm is O(N) in serial computation and O(log N) in parallel computation on O(N) processors for most practical kinematic chains. The idea is to assemble a kinematic chain by adding the joints one by one and compute the constraint forces at the new joints using the principle of virtual work. The parallelism of the algorithm can be adapted for parallel processing systems with any number of processors by simply changing the assembly order. Simulation examples on an 8-node cluster demonstrate the effectiveness of the algorithm.

Presents an interface for creating whole-body motions of human and animal characters without reference motion. Its basic function is to enable animators to generate a natural motion by dragging a link to an arbitrary position with any number of links pinned in the global frame, as well as other constraints such as desired joint angles and joint motion ranges. Each constraint can be switched on or off, strengthened or weakened for each joint at a user's will. The interface is based on an online inverse kinematics technique that allows more flexible attachment of pins and various types of constraints. Editing or retargeting captured motion requires only a small modification to the original method, although the method can create natural motions from scratch. We also demonstrate the power and usability of the proposed method by a number of example motion clips.

Behavior recognition process and behavior generation process have a close relationship in humans' brains. It is expected that humans' brains understand the meaning of behavior and create symbols through co-development of recognition and generation processes. In this paper, we propose a novel method for the integration of behavior patterns and symbols using associative memory in order to realize the co-development processing. In the model, behavior recognition process and generation process are practiced based on a mutual dynamics. We also confirmed the feasibility of the method on humanoid simulator.

Since humanoid robots have a number of degrees-of-freedom in general, a pattern-based approach of the motion control reduces its difficulty. It is necessary, however, to absorb and compensate disturbances in order to maintain the stability of robots in the real world. We developed a balancing method for humanoid robots with a little modification of predesigned motion trajectories. The method proposed has an advantage that it is allowed to choose any combination of joints as modified properties, so that it has enough flexibility, being applicable for various types of robots and motions. It consists of two phases; in the first phase, the referential COG displacement is decided in accordance with both the short-term and the long-term absorption of disturbances. And in the second phase, the COG is manipulated with the whole-body cooperation, using the COG Jacobian. We verified the validity of the method with some simulations.

Precise measurements of geometry should accompany robotic equipment in operating theatres, for greatest advantage. For deforming organs, including liver, intraoperative geometric measurements play an essential role in computer surgery in addition to pre-operative geometric information from CT, MRI and so on. We developed a laser-pointing endoscope using an optical galvano scanner and a 955 fps high-speed camera. The laser-pointing endoscope system acquires and visualizes the shape of the area of interest in a flash of time. Applications of the system also include the touch screen interface for nonmaster-slave operation of surgical robots, where the 3D coordinates of the touched point on screen are measured by the system and guide a robot. Results of in-vivo experiments on a liver of pig verify the effectiveness of the proposed system.

Since a biped humanoid inherently suffers from instability and always risks tipping over, ensuring high stability and reliability of walking is one of the most important goals. The paper proposes a walk control consisting of a feedforward dynamic pattern and a feedback sensory reflex. The dynamic pattern is a rhythmic and periodic motion, which satisfies the constraints of dynamic stability and ground conditions, and is generated assuming that the models of the humanoid and the environment are known. The sensory reflex is a simple, but rapid motion programmed with respect to sensory information. The sensory reflex, we propose, consists of the body posture control, the actual ZMP (zero moment point) control, and the landing time control. With the dynamic pattern and the sensory reflex, it is possible for the humanoid to walk rhythmically and to adapt itself to environmental uncertainties. The effectiveness of our proposed method was confirmed by walk experiments of an actual 26 DOF humanoid on an unknown rough terrain and in the presence of disturbances.

Mimesis is a primitive learning framework and origins of human intelligence. We have developed a behavior acquisition and understanding system based on the mimesis. This system is able to abstract observed others' behaviors into conceptual symbols, to recognize others' behavior using the primitive symbols, and to generate self motion patterns using the primitive symbols. In this paper, we mention the integration of mimesis loop which is the acquisition and development system based on mimesis, and confirmation of the feasibility against whole body motions on virtual humanoids.

Robotics can serve as a testbed for cognitive theories. One behavioral criterion for comparing theories is the extent to which their implementations can learn to exploit new environmental opportunities. Furthermore, a robotics testbed forces researcher to confront fundamental issues concerning how internal representations are grounded in activity. In our approach, a mobile robot takes the role of a creature that must survive in an unknown environment. The robot has no a priori knowledge about what constitutes a suitable goal-what is edible, inedible, or dangerous-or even its shape or how its body works. Nevertheless, the robot learns how to survive. The robot does this by tracking segmented regions of its camera image while moving. The robot projects these regions into a canonical wavelet domain that highlights color and intensity changes at various scales. This reveals sensory invariance that is readily extracted with Bayesian statistics. The robot simultaneously learns an adaptable sensorimotor mapping by recording how motor signals transform the locations of regions on its camera image. The robot learn about its own physical extension when it touches an object. But it also undergoes an internal state change analogous to the thirst quenching or nausea producing effects of intake in animals. This allows the robot to learn what an object affords by relating these effects to learned clusters of invariance. In this way primitive symbols emerge. These protosymbols provide the robot with goals that it can achieve by using its sensorimotor mapping to navigate, for example, toward food and away from danger.

Minimally invasive direct coronary artery bypass (MIDCAB) requires of surgeons precision of hand skill and mental concentration, since it needs to work on beating hearts. We propose a surgical robot system that compensates motions of organs during operations. The motion canceling robot system consists of three technologies; visual synchronization, motion synchronization and master-slave control. The visual stabilization provides the surgeon with the image of stabilized target point on the video monitor. The surgeon operates the master robot referring to the stabilized image. The motion stabilization, on the other hand, controls the slave robot being synchronized with the heart beat, which is the function of the master-slave control. Master-slave transforms the master motion and controls the slave robot. In this paper, we verify the effectiveness of the prototype system by in-vivo experiment.

Design of mechanical compliance would be one of the most important technical foci in making humanoid robots really interactive with the humans. For safety insurance mechanical compliance should be developed for humanoid robots. The introduction of the passive compliance to humanoid robots has large possibility for achieving human skill by using the dynamical energy stored in the compliant members. Programmable passive compliance plays an important role in coping with the changing environments and task execution. We evaluate the effectiveness of the passive compliance for the realization of human skill and design a programmable passive compliance mechanism 'PPC cybernetic shoulder' which is a four degree of freedom shoulder mechanism for humanoid robots using a closed kinematic chain. The programmability of the PPC cybernetic shoulder is evaluated by experiments.

We study the behavior control of a robot using orbits of nonlinear dynamics. The behavior generation and control using entrainment and synchronization phenomena in nonlinear dynamical systems is discussed. The behavior of the Arnold equation, which is known to show the chaotic behavior of non-compressive perfect fluid, is analyzed. Methods to integrate the dynamics into the information processing system of a robot is discussed.

Presents a practical motion planner for humanoids and animated human figures. Modeling human motions as a sum of rigid body and cyclic motions, we identify body postures that represent the rigid-body part of typical motion patterns. This leads to a model of the configuration space that consists of a multi-layered grid, each layer corresponding to a single posture. A global search through this reduced configuration space yields a feasible path and the corresponding postures along the path. A velocity profile is calculated along the optimal path, subject to the speed and acceleration limits assumed for each posture. Cyclic motions, generated from "primitive" cyclic motion patterns for each posture, are then added to the trajectory produced by the path planner. This "kinematic" motion is then modified by a dynamics filter to result in dynamically consistent behavior. Examples are presented which demonstrate the use of this planner in an office environment.

This paper describes an efficient algorithm for the forward dynamics of open kinematic chains with O(N) complexity, where N is the number of links in the chain. The method is based on the principle of virtual work and does not use any theory in linear algebra or the concept of articulated body inertia. The idea of this method is to add a link one by one from the leaflinks to the root evaluating the constraint force at each new joint. The algorithm consists of two iterative procedures: from the leaflinks to the root to compute the constraint forces, and from the root to the leaf to compute the joint accelerations. Some numerical examples show the efficiency of the proposed algorithm. Similarity and differences with other O(N) algorithms are also discussed.

The reliability of stable walk and the development of high performance components are two crucial issues to develop a humanoid with human-like physical capability. In order to for the humanoid to walk smoothly and to adapt to unknown environments, we first propose a balance control that combines a feedforward dynamic pattern and a feedback sensory reflection. Then, we present a method for clarifying the relationship between the physical capability and actuator's specifications. Using this method, it is possible to predict the walking speed based on known actuator specifications and to obtain the necessary specifications to accomplish a desired walking speed. Finally, experiments of an 26-DOF humanoid and simulation examples are provided to illustrate the effectiveness of the proposed method.

This paper presents a virtual humanoid robot platform (V-HRP for short) on which we can develop the identical controller for a virtual humanoid robot and its real counterpart. The unification of the controllers for the virtual and real robot has been realized by introducing software adapters for two robots respectively and employing ART-Linux on which real-time processing is available at the user level. Thanks to the unification, the controllers can share softwares with the dynamics simulator of V-HRP, including the parameter parser, kinematics and dynamics computations and the collision detector. This feature can make the development of the controllers more efficient and the developed controllers more reliable.

Nonholonomic constraints are exploited to design a controllable n-joint manipulator with only two inputs. Gears subject to nonholonomic constraints are designed to transmit velocities from the inputs to the passive joints. The system possesses a triangular structure for which a conversion into chained form is presented. The nonholonomic manipulator can, therefore, be controlled with an open loop or a closed loop using existing controllers for chained form. Mechanical design is established, and experimental results proved the usefulness of design of the nonholonomic manipulator and applied control schemes. While previous publications have assumed that the nonholonomic systems are given and have developed theory for these systems, this paper points out a new direction where the nonholonomic theory is used to design controllable and stabilizable systems.

A wheeled mobile robot with trailers has been studied as a class of nonholonomic systems. It is proved that a system of a tractor and trailers with an appropriate connecting mechanism can be stabilized to desired positions via nonholonomic motion control. Trailers, on the other hand, have been developed and widely used in the industry. The main focus of industrial design is set on reducing tracking error from a reference trajectory. This paper attempts to bridge over the gap between these two approaches. We develop a design theory of trailer systems with passive steering. The designed systems show a good performance in practical path following, and accept the chained form transformation and nonlinear control strategies for nonholonomic systems.

In this paper, we develop a method to impart the chaotic nature to a mobile robot. The chaotic mobile robot implies a mobile robot with a controller that ensures chaotic motions. Chaotic motion is characterized by the topological transitivity and the sensitive dependence on initial conditions. Due to the topological transitivity, the chaotic mobile robot is guaranteed to scan the whole connected workspace. For scanning motion, the chaotic robot neither requires the map of the workspace nor plans the global motion. It only requires the measurement of the local normal of the workspace boundary when it comes close to it. We design the controller such that the total dynamics of the mobile robot is represented by the Arnold equation, which is known to show the chaotic behavior of noncompressive perfect fluid. Experimental results and their analysis illustrate the usefulness of the proposed controller.

A wheeled mobile robot with trailers has been studied as a class of nonholonomic systems. It is proved that a system of a tractor and trailers with an appropriate connecting mechanism can be stabilized to desired positions via nonholonomic motion control. Trailers, on the other hand, have been developed and widely used in the industry. The main focus of industrial design is set on reducing tracking error from a reference trajectory. This paper attempts to bridge over the gap between these two approaches. We develop a design theory of trailer systems with passive steering. The designed systems show a good performance in practical path following, and accept the chained form transformation and nonlinear control strategies for nonholonomic systems.

Design and control of mechanical compliance would be one of the most important technical foci in making humanoid robots really interactive with humans. For task execution and safety insurance the issue must be discussed and offers useful and realistic solutions. We propose a theoretical design principle of mechanical compliance. Passive compliance implies a mechanically embedded compliance in drive systems and is reliable but not tunable in nature, while active compliance is a controlled compliance and, therefore, widely tunable, but less reliable specially in the high frequency domain. The basic idea of the paper is to use active compliance in the lower frequency domain and to rely on passive compliance in the higher frequency. H/sub /spl infin// control theory based on systems identification allows a systematic method to design the hybrid compliance in frequency domain. The proposed design is applied to the shoulder mechanism of a humanoid torso robot. Its implementation and experiments are to be shown with successful results.

Humanoid robots are required to make a variety of dynamics and even expressive motions in changing environments. However, the conventional methods for generating humanoid motions fail do achieve this requirement since they can only generate quite artificial and predefined motions through rather complicated optimization processes. In this paper, we propose the concept of "dynamics filter" which transforms a physically inconsistent motion into a consistent one, and provide an example of its implementation using feedback control and local optimization. The optimization is based on the equation of motion of constrained kinematic chains, which is derived from our previously proposed method for computing the dynamics of structure-varying kinematic chains. The proposed method can be applied to online motion generator of humanoid robots.

Since a biped humanoid inherently suffers from instability and always risks to tipping over, stable and reliable biped walking is the most important goal. The simulator is a significant tool to pursue this goal. In this paper, we first present a method for constructing a humanoid simulator that can closely model and predict the motion of an actual humanoid. We then propose a balance controller consisting of an off-line walk-pattern generator and a real-time modification. Using the simulator, we can predict the humanoid's physical capability subject to the constraints of actuators, and clarify the required specifications of actuators to execute a desired task. The functions of the developed simulator and the effectiveness of the proposed balance controller were evaluated through simulated walks on an unknown rough terrain, soft ground, and an environment in the presence of disturbances.

A new type of mobility is discussed for space projects such as the MUSES-C aiming at small asteroid exploration. We propose the use of electro-magnetic levitation in order to integrate a mobility into the microgravity rover. The rover has a spherical shape and a smaller spherical shell inside. Four electromagnets are symmetrically located between the outer sphere surface and the inner sphere shell with one end of each directed to the center of the shell. With electromagnetic force of the magnets, a sphere iron ball inside the shell is controlled and levitated. When the rover lifts the ball inside with the electro-magnetic force, the rover is in return pressed down the ground by the reaction force, due to which the rover system not only gains upward momentum for floatation, but also obtains friction that enables its rolling on the ground. The prototype microgravity rover was developed and experimental results indicate effectiveness of the proposed mobility.

Manipulators with free joints are second-order nonholonomic systems whose dynamical constraints are nonintegrable. Such systems are known to be under-actuated systems which can be controlled by less actuators than the dimension of the configuration space. Previously proposed methods to control free-joint manipulators have been based on the assumption of perfectly frictionless free joints. In this paper, averaging analysis of 3R manipulators with one actuator clarifies that their frictionless models are Hamiltonian systems with conservation of an energy-like quantity. From experiments and simulations, 3R free-joint manipulators with friction at the free joints show dissipative behaviors converging to an equilibrium point. A convergence control method using the energy-like quantity to stabilize to the equilibrium point is proposed.

This paper discusses the dynamics computation of structure-varying kinematic chains which imply mechanical link systems whose structure may change from open kinematic chain to closed one and vice versa. The proposed algorithm can handle and compute the dynamics and motions of any rigid link systems in a seamless manner without switching among algorithms. The computation is developed on the foundation of the dynamics computation algorithms established in robotics, which is superior in efficiency due to explicit use of the generalized coordinates to those used in the general-purpose motion analysis softwares. Although the structure-varying kinematic chains are commonly found in computing human and animal motions, the computation of their dynamics has not been discussed in literature. The developed computation will provide a general algorithm for the computation of motion and control of humanoid robots and computer graphics human figures.

This work presents a study of the human/humanoid locomotion system and a method for adaptation of the human motion capture data (HMCD) for driving a humanoid robot. The analysis uses the previously defined concept of the zero moment point (ZMP) which provides a basis for the adaptation of the HMCD. An appropriate model of the robot foot, in agreement with the HMCD, is proposed. This model is used to plan a desired ZMP trajectory. A scheme for approximately matching the actual ZMP trajectory to the desired ZMP trajectory, through periodic joint motion correction at selected joints, is discussed. A method for resolving the ground reaction forces at the foot is also proposed.

The integration theory of reactive behaviours is discussed. A linear emerging model is adopted where the motion of a robot is represented as the weighted linear sum of reactive behaviours. The weights are defined as differentiable nonlinear functions of sensor signals and parameters. We propose approaches toward skill learning and skill abstraction based on the sensor space model, where the parameters are systematically tuned through iteration of trials such that the sensor signals converge to the given teacher signals. The learning algorithm and the abstraction algorithm are experimentally applied to the reactive grasp of a three-fingered robot hand. The experimental results illustrate the effectiveness of the proposed algorithms.

Discusses the dynamics computation of structure-varying kinematic chains which imply mechanical link systems whose structure may change from open kinematic chain to closed one and vice versa. The proposed algorithm can handle structure changes in a seamless manner without switching among algorithms for different kinematic chains. The structure-varying kinematic chains are commonly found in computing human motions. The developed computation will provide the general algorithm for the computation of motion and control of humanoid robots and computer graphic human figures.

This paper proposes an approach for generating a physically consistent trajectory from a few key frames of flight motion of a human-like multibody system. Systems in flight are subject to angular momentum conservation, a nonholonomic constraint, and linear momentum conservation. In the proposed approach, motion of one of the links is first generated independently. The motion of all other links is planned appropriately to satisfy the constraints of motion. This work has potential applications ranging from designing dynamic animations to planning free flight motion for humanoid robots.

Discusses the integration of "mechanical softness" into the humanoid robot mechanisms design. The mechanical softness includes such requirements as human-like high mobility and human-like sensitive compliance. We focus on the shoulder mechanism, and propose a parallel mechanism to integrate the two requirements. The mechanism is called the cybernetic shoulder and possesses three degrees-of-freedom. The nature of motion curves of the human shoulder, and the design and development of the cybernetic shoulder are described with the computation issue of kinematics. The integration of compliance into the parallel mechanism is also discussed, and its experimental evaluation is made.

In this paper, we develop a method to impart the chaotic nature to a mobile robot. The chaotic mobile robot implies a mobile robot with a controller that ensures chaotic motions. Chaotic motion is characterized by the topological transitivity and the sensitive dependence on initial conditions. Due to the topological transitivity, the chaotic mobile robot is guaranteed to scan the whole connected workspace. For scanning motion, the chaotic robot does not require the map of workspace. It only requires to measure the local normal of the workspace boundary when it comes close to it. We design the controller such that the total dynamics of mobile robot is represented by the Arnold equation, which is known to show the chaotic behavior of noncompressive perfect fluid. Experimental results and their analysis illustrate the usefulness of the proposed controller.

This paper discusses the dynamics computation of closed kinematic chains, especially those found in motions of human figures. A number of efficient dynamics computation algorithms have been established in robotics for open kinematic chains and particular types of closed kinematic chains such as parallel five-bar link mechanisms and the Stewart platform. The dynamics computation of closed kinematic chains, however is still challenging and among open research issues. In this paper, we describe the mobility of closed kinematic chains by the minimal set of independent variables, which we call the generalized coordinates of a closed kinematic chain. We then develop a systematic procedure to find them out, and establish the computational algorithms for the inverse and forward dynamics of any closed kinematic chains. The numerical examples show the effectiveness of the algorithms in particular for computing high-degrees-of-freedom human/animal motions.

We propose a control scheme of parallel manipulators focusing on the accuracy of acceleration on the endplate, which is an important factor when parallel manipulators are used as acceleration displays. We use two controllers-a dynamic controller to achieve accuracy of position and to stabilize the system, and an H/sub /spl infin// controller to feedback the acceleration measured on the endplate. The main problem of dynamic control is computational complexity. In order to reduce computation time for inverse dynamics, a parallel processing method called multi-thread programming is applied. The H/sub /spl infin// controller is added outside the closed loop of the dynamic control to remove the vibration of the structure and the influence of modeling errors in the dynamic controller.

Exploiting unique features of nonholonomic systems, innovative and advantageous mechanisms can be designed. We have proposed the chained form manipulator which is a controllable n-joint manipulator with only two actuators. The chained form manipulator is designed not only to satisfy chained form convertibility, but also to achieve control simplicity. Design requirements and mechanical design of the chained form manipulator were proposed in our prior work. For the experimental verifications, we fabricated a prototype. So far, various control strategies for nonholonomic systems have been proposed. In this paper, an efficient motion planning scheme for the chained form is presented to approximate any holonomic path with the feasible nonholonomic path. Furthermore, a new concept of motion planning is proposed through the analysis of the initial-condition sensitivity. Combining these two approaches, the motion planning scheme is constructed towards practical applications. Presented experimental results show the usefulness of the design and the applied control scheme.

We propose an experimental system of a space robot in the microgravity environment at the Japan Microgravity Center where microgravity (less than 10/sup -5/ g for 10 sec) is generated by a free-fall of 490 m. In the environment, we performed experiments for a visual feedback control of a space robot capturing a target. To measure the position of the target and the motion of the robot, two CCD cameras are used and their images are processed by the tracking vision. After the introduction of the experimental system, we establish two computational methods to identify the 3D position and orientation of the robot base, and show the experiments and their results to evaluate the proposed methods.

We propose a free-flying space robot whose center-of-mass is fixed to the base body and invariant to changes of the configuration. The space robot with such a structure, named 'center-of-mass invariant structure,' has the following characteristics: (1) an experimental system on the ground is simply built, (2) the computational cost of the generalized Jacobian matrix is reduced, and (3) motions in 2D planes are holonomic. We developed a prototype of the space robot with the center-of-mass invariant structure. The result of a preliminary experiment is to be shown.

Exploiting the unique features of nonholonomic systems, we have proposed and fabricated the nonholonomic manipulator which is a controllable n joint manipulator with only two actuators. In order to create its nonholonomic constraint, a special type of velocity transmission, called the nonholonomic gear, was used. There are many possible alternatives of designing underactuated manipulators using the nonholonomic gear. The nonholonomic manipulator was designed focusing on the mechanical simplicity. In this paper, we establish the design procedure of underactuated manipulators whose kinematic model can be converted to the chained form, giving priority to the practical control aspects. As an example that has control simplicity, the chained form manipulator was designed according to the proposed design concept.

A manipulator with free joints is a class of under-actuated mechanisms. Control of such systems is one of major topics in robotics and control engineering. In this paper, we apply the averaging method to manipulators connected by free joints and describe their behaviors in response to periodic inputs. We analyze 2R and 3R free-joint manipulators and show that each system has invariant manifolds defined by an energy-like conserved quantity. We also develop a control method via feedback modulation of the input amplitude to reach a desired invariant manifold. The effectiveness of the method is verified by computer simulations.

We have established a research collaboration project to develop a precise and high-speed arm applied to assembly process in electrical product and machine industries, supported by the International Robots and Factory Automation Foundation (IROFA). The paper addresses the background of the development, the required specification for the robot arm, and the brief introduction of the prototype arm.

The nonlinear motion of underactuated mechanisms has drawn recent interests of researchers. Underactuated mechanical systems are often subject to so-called nonholonomic constraints which are related to many both theoretical and practical issues. We first analyze the nonlinear behavior of a two-joint planar manipulator with the second joint free, from nonlinear dynamics point of view. We then discuss the simultaneous positioning of both joints. We use a time-periodic input and propose an amplitude modulation of the feedback error. The analysis via the Poincare map shows that the behavior becomes chaotic with large amplitude. The effectiveness of the proposed positioning control is verified by experiments.

Nonholonomic system is now one of the major topics in robotics. In this paper, we discuss the nonlinear behavior of a two-joint planer manipulator with the second joint free for time-periodic inputs. First, we illustrate that when the amplitude remains small, the Poincare map of the system follows an ellipse-like closed path and it becomes chaotic with large amplitude. We then propose a control method to position the both joints via amplitude modulation of the position error. The effectiveness and robustness of the control method are verified by experiments.

A free-flying space robot with a 6-DOF manipulator cannot follow an arbitrary trajectory in the 9D generalized coordinates (3 of the satellite orientation and 6 of the manipulator) with only manipulator joint control, though it was shown to be commonly controllable in the literature. In this paper, we propose a method to approximate the desired 9D path by introducing a perturbation around it. We call the approximated trajectory the "spiral motion". A computational scheme for planning the spiral motion is presented, and is followed by computer simulation that illustrates the effectiveness of the scheme.

We proposed a nonholonomic manipulator which is a controllable n joints manipulator with only two inputs, exploiting a special kind of velocity transmission called a nonholonomic gear. Since the nonholonomic manipulator was theoretically designed from the viewpoint of kinematic constraints and nonlinear control, the mechanical implementation and prototyping are extremely important in practice. In this paper, the principle of mechanical design of a nonholonomic manipulator is established, and the experimental results are shown using a prototype nonholonomic manipulator.

In laparoscopic surgery, forceps are stuck through trocars into the abdominal cavity, and dissect and grasp internal organs. Because of their straight shapes, the current forceps suffer from the narrow range of operation and limit the skill of surgeons. In this paper, we propose active forceps actuated by unconventional use of shape memory alloy. In developing active forceps, we utilize two properties of shape memory alloy, shape memory effect and super elasticity The forceps possess both stiffness for transmitting operation forces from surgeons and flexibility in changing their shapes.

Nonholonomic systems are typically controllable in a configuration space of higher dimension than the input space. Here, it is shown how nonholonomic constraints can be exploited to design a controllable n-joint manipulator with only two inputs. Gears subject to nonholonomic constraints are designed to transmit velocities from the inputs to the unactuated joints. The designed nonholonomic manipulator is shown to be completely controllable in the whole configuration space. The system is designed with a triangular structure for which a conversion into chained form is presented. The nonholonomic manipulator can, therefore, be controlled using existing controllers for chained form.>

Power grasp is redefined as a type of grasp that its mechanism can resist passively against external forces without relying on feedback control of joint torques. A computational algorithm is invented to calculate the critical external force, a force which is requisite to move the grasped object in a definite direction. Virtual work is proposed as the quality measure of the robustness of power grasp. Because this measure is a scalar, it is convenient and suitable for the planning of power grasp. The effectiveness of the computational algorithm of critical external force and quality measure of robustness of power grasp is verified with a numerical example.>

As a large-scale space structure, the authors propose a space multibody structure that consists of many relatively small bodies mutually connected by passive joints. They assume that one of the bodies is equipped with an orientation-control device, such as control momentum wheels. Being free from elasticity-induced vibration is an advantage of the structure. The authors analyze the dynamics of nonholonomic behavior of the structure and propose a control scheme to stabilize its shape.

Nonholonomic redundancy is an intrinsic property of nonholonomic mechanical systems. A free-flying space robot is a nonholonomic mechanical system, and exhibits the presence of nonholonomic redundancy even in the absence of ordinary kinematic redundancy. Like ordinary kinematic redundancy, nonholonomic redundancy can also be utilized while planning trajectories for the system. In the paper, a trajectory planning scheme for a 6-DOF space robot is developed in which nonholonomic redundancy for avoiding joint limits and obstacles is utilized.>

A free-flying space robot has kinematic and dynamic features different from those fixed on the Earth because of the momentum constraints that govern its motion. The authors present the solution to the inverse dynamics problem of a space robotic system in the presence of external generalized forces. While solving for the inverse dynamics, the computations for the inverse kinematics are considered simultaneously, and both computations are developed on the basis of momentum constraints. An efficient computational scheme for the inverse dynamics problem is then established. Space robotic systems have an intrinsic feature that can be utilized to reduce the computational time by parallel recursion. Finally, the scope of parallel recursion is discussed.>

Discusses 3D motion of underwater vehicles. The authors describe kinematics of an underwater vehicle by six state variables and four inputs, and use a Lyapunov-like function to develop a nonlinear tracking control scheme. The control method effectively makes use of the nonholonomic nature of the system. Simulation results agreed with the theoretical predictions and confirmed the usefulness of the proposed scheme.>

A free-flying space robot for the construction and maintenance of space structures is considered. Such a robotic system has kinematic and dynamic features that differ from those fixed on the earth mainly because of the momentum constraints that govern its motion. The solution to the inverse dynamics problem of a space robotic system in the presence of external generalized forces is presented. The computations for the inverse kinematics are considered simultaneously, and both computations are developed on the basis of momentum constraints. An efficient computational scheme for the inverse dynamics problem is then established. An intrinsic feature of space robotic systems that can be utilized to reduce the computational time by parallel recursion is discussed, as is the importance of the role of momentum constraints in the solution of inverse dynamics.>

A general and systematic method to find the dynamic parameters, known as the base parameters, that directly contribute to joint torques for both open and closed kinematic chains is presented. The principal base parameters are defined as a set of the base parameters with their order of sensitivity. The principal base parameters are obtained by taking into account the nonlinearity. It is also shown that for a closed kinematic chain a set of base parameters are invariant to the location and the number of the actuated joints if they can sufficiently activate the mechanism.>

A force-sensitive multi-DOF (degree-of-freedom) wrist suitable for force and collision control is presented. Similar kinds of wrists have been developed before but they suffered from insufficiency of power or massiveness. Having practical applications in mind, the authors present a magnetically driven system in which the nonlinear and limited electromagnetic force can be used effectively and the output force can be applied as uniformly as possible in all directions. The developed wrist can drive a hand or a tool within the range of +or-1 mm in all three directions and can apply up to 100-N of force anywhere in the work space. The wrist is clean due to the absence of mechanical friction.>

A submersible can carry limited supply of fuel onboard. Use of control planes along with propeller for major part of its motion can conserve much energy; thus extending the duration of underwater missions. Such an underwater vehicle has nonholomic nature due to its nonlinear kinematic structure. The authors describe the kinematics of an underwater vehicle by six state variables and four inputs, and use a Lyapunov-like function to develop a nonlinear feedback control scheme. Effectiveness of the nonlinear control scheme was verified through numerical simulation.>

The control problem for robot manipulators having some unactuated joints is addressed. The nonholonomic nature of the constraint expressing the dynamics of the free joints is recognized in the general case, and conditions are derived to identify special cases in which such a constraint is integrable. In contrast to most examples in the literature, the free-joint dynamics is an instance of second-order nonholonomic constraint. It is shown that smooth feedback stabilization to a single equilibrium point is not possible. A feedback scheme achieving stabilization to a manifold of equilibrium positions is proposed. Its correctness is established theoretically as well as confirmed by simulation results.>

The path planning of nonholonomic motion of space robot systems is discussed. A space vehicle with a 6-DOF (degrees of freedom) manipulator is described as a nine-variable system with six inputs. It is shown that, by carefully utilizing the nonholonomic mechanical structure, the vehicle orientation in addition to the joint variables of the manipulator can be controlled by actuating only the joint variables. The nonholonomic mechanical structure of space robot systems is shown. A rigorous mathematical proof of the nonholonomic nature of the free-flying space robot systems is provided using Frobenius's theorem. A method for nonholonomic motion planning for space robot systems is established by using a Lyapunov function.>

The nonholonomic path planning of space robot systems is discussed. A space vehicle with a 6-DOF (degree of freedom) manipulator is described as a nine-variable system with six inputs. Utilizing the nonholonomic nature, the vehicle orientation can be controlled in addition to the joint variables of the manipulator by actuating only the joint variables, if the trajectory is carefully planned. A path-planning method of nonholonomic motion is developed using a Lyapunov function.>

A singularity-free parameterization is proposed that constantly allows full utilization of the actuation redundancy of a closed-link mechanism. The improvement due to the parameterization is numerically demonstrated. The advantage of actuation redundancy is investigated in the light of performance index minimization and maximum payload. The optimization of actuation redundancy is shown to result in significant improvement in both respects.>

The authors discuss the nonholonomic mechanical structure of a space robot and its path planning. The conservation of angular momentum works as a nonholonomic constraint, whereas the conservation of linear momentum is a holonomic one. In this framework a vehicle with a 6-DOF (six-degree-of-freedom) manipulator is described as a nine-variable system with six inputs. This implies the possibility of controlling the vehicle orientation as well as the joint variables of the manipulator by actuating the joint variables only if the trajectory is carefully planned, although both the variables and the trajectory cannot be controlled independently. In planning a feasible path, a system than consists of a vehicle and a 6-DOF manipulator can be treated as a 9-DOF kinematically redundant system. The nonholonomic mechanical structure of the space vehicle/manipulator system is shown, and a path planning scheme for nonholonomic systems using Lyapunov functions is proposed.>

A general statistical fusion method motivated by the geometry of uncertainties is proposed for robotic systems with multiple sensors. The treatment of nonlinearity is generalized so as to include both the structural nonlinearity and the computational nonlinearity. First, assuming Gaussian noise additive to the sensory data, the uncertainty ellipsoid associated with the covariance matrix of the error of the sensory information is defined. Second, the optimal fusion is defined as the one, among all the possible linear combinations of sensory information, that minimizes the geometrical volume of the ellipsoid. The resultant fusion equation coincides with those obtained by Bayesian inference, Kalman filter theory, and the weighted least-squares estimation. Finally, the method is extended to include the fusion of partial information.>

The authors discuss a general and systematic computational scheme of the inverse dynamics of closed-link mechanisms. It is derived by using d'Alembert's principle and obtained without computing the Lagrange Multipliers. To account for the constraints, only the Jacobian matrix of the passive joint angles in terms of actuated ones is required. Given a nonredundant actuator system, this allows a unique representation of the constraints even for complicated multiloop closed-link mechanisms. The inverse dynamics of closed-link mechanisms that contain redundant actuators and their redundancy optimization are also discussed. For a redundant actuation system that contains N/sub r/ redundant actuators, the passive joint angles are represented by N/sub r/+1 independent ways as functions of actuated joints. Using their Jacobian matrices, the actuation redundancy of a closed-link mechanism is parameterized by an N/sub r/-dimensional arbitrary vector in a linear equation. Numerical examples are given to show the computational efficiency of inverse dynamics computation and the potential of closed-link manipulators with actuation redundancy.>

A general and systematic scheme for computing the dynamics of closed link mechanisms is derived using D'Alembert's principle. To account for the constraints, only the Jacobian matrix of the function which represents the passive joint angles in terms of the actuated ones is required. Given a nonredundant actuator system, this allows a unique representation of the constraints, even for a complicated multi-loop closed link mechanism. The scheme is computationally efficient because it is not necessary to compute Lagrange multipliers. A redundant actuator system is formulated, and the utilization of redundancy is actuation is illustrated.>

This paper discusses the mechanics of coordinative manipulation by multiple robot manipulators or a multifingered robot hand. The coordinative manipulation problem is divided into two phases. One is determining the resultant force by multiple robotic mechanisms, and the other is determining the internal force between them. The resultant force is used for the manipulation of an object subjected by the external forces or the environmental constraints. The internal force is used for adapting the robotic mechanisms to uncertainty and variety of the static friction. A dynamic coordinative control scheme is proposed for determining the resultant forces. The optimal internal force is defined as the internal force that yields the minimal norm force satisfying static frictional constraints. The optimal internal force promotes the stability of prehension, while the conventional method sometimes result in too much internal force and reduce the stability. Finally, by applying a non-linear programming method, it is clarified that the optimal solution is necessarily obtained by solving, at most, 2(2m-1) (m is the number of robotic mechanisms) sets of algebraic equations if it exists.