Yui Endo

Humanoid robot as an evaluator of assistive devices

This paper presents a basic study on feasibility of usage of humanoid robots as an evaluator of assistive devices, by taking advantage of its anthropomorphic shape. In this new application humanoid are expected to help evaluation through quantitative measures, which is difficult with human subjects, and also to reduce the burden coming from ethical concerns with costly tests by human subjects. Taking a passive supportive wear “Smart Suit Lite” designed to relieve the load at lower back as an example, we have conducted pilot experiments by using the humanoid robot HRP-4C. The motion to be performed by the humanoid is obtained through retargeting technique from measured human lifting motion. The supportive effect is first estimated by simulation taking into account the mechanism of the supportive device. The experimentation of humanoid hardware brought us encouraging results on the basic feasibility of this application, as we observed a clear decrease of the torque for lifting when wearing the device as expected by the simulation.

Optimization-Based Grasp Posture Generation Method of Digital Hand for Virtual Ergonomics Assessment

Automatically generating humanlike grasp postures of the digital hand is a key issue for the virtual ergonomic assessment of the industrial products. In this paper we propose a new optimization-based approach for generating the realistic grasp posture. As an objective function, we use the number of the contact points, the fit of the specific part of the hand surface for the feature edges of the product surface and the margin for the constraints on the joint angle limits of the figures. The experimental studies on the grasp posture generation for the digital camera indicate that more realistic grasp posture could be generated using the proposed optimization-based method than the one using our former method.

Reconstructing individual hand models from motion capture data

In this paper, we propose a new method of reconstructing the hand models for individuals, which include the link structure models, the homologous skin surface models and the homologous tetrahedral mesh models in a reference posture. As for the link structure model, the local coordinate system related to each link consists of the joint rotation center and the axes of joint rotation, which can be estimated based on the trajectories of optimal markers on the relative skin surface region of the subject obtained from the motion capture system. The skin surface model is defined as a three-dimensional triangular mesh, obtained by deforming a template mesh so as to fit the landmark vertices to the relative marker positions obtained motion capture system. In this process, anatomical dimensions for the subject, manually measured by a caliper, are also used as the deformation constraints.

Virtual Ergonomic Assessment on Handheld Products based on Virtual Grasping by Digital Hand

The purpose of this research is to develop a system for virtual ergonomic assessment of products without real subjects and physical mockups by integrating a digital hand model with a product model. In previous work, we developed functions of a semi-automatic grasp planning for the digital hand and of quantitatively evaluating the grasp stability of the product based on the force-closure and the grasp quality in our system. We also confirmed the validity of the results from these functions by comparing them with the real grasp postures. However, only evaluating the grasp stability could not necessarily derive the appropriate grasp postures. To solve this problem, in this paper, we propose a new function of evaluating “ease of grasping (EOG)” for the grasp posture based on EOG-map constructed from principal component analysis for finger joint angles in real subjects’ grasps. We also developed another function of optimizing the grasp posture to avoid inappropriate postures by evaluating the posture similarity on EOGmap in the system.

Anatomographic Volumetric Skin-Musculoskeletal Model and Its Kinematic Deformation With Surface-Based SSD

Conventional musculoskeletal models mainly consist of bones modeled by rigid linkages, and muscles, tendons, and ligaments modeled by ideal wires. The lack of volumetric modeling of muscle makes a representation of interaction between muscles and natural muscle pathway difficult. This difficulty results in a physiologically inappropriate estimation of a muscle momentum arm that is critical for a muscle activity estimation. In this letter, we develop a volumetric skin-musculoskeletal model based on an anatomographic human shape database to improve an estimation of a muscle momentum arm. A volumetric deformation of surface skin and muscle is realized by an extended skeleton subspace deformation (SSD, the linear blend skinning algorithm) that considers a surface profile of bone with low computational cost. This extended SSD considers a sub-bone that is projected to the bone surface polygon so that the skin and muscle deformation is significantly affected by the bone surface profile. The surface-based SSD realized the natural skin deformation avoiding a penetration between skin and bones during trunk rotation that results in a physiologically appropriate estimation of a muscle momentum arm. The volumetric skin-musculoskeletal model and the surface-based SSD estimates the momentum arm of vastus lateralis with 14.1% maximum error from a literature values, though there is 44.8% maximum error with the wire musculoskeletal model. This model would accurize a muscle activity estimation that leads to a more correct understanding of human motion control/generation mechanisms.

Estimation of Arbitrary Human Models from Anthropometric Dimensions

In this paper, we describe a novel approach for reconstructing arbitrary whole-body human models from an arbitrary sparse subset of anthropometric dimensions. Firstly, a comprehensive set of dimensions is estimated from the subset via the principal component space for the dimensions. Then, a skin surface model with the obtained comprehensive set of dimensions is constructed by deforming a whole-body human model template. The result is validated based on the error distribution of the dimensions of the obtained surface mesh for the target.

Ingress and egress motion strategies of elderly and young passengers for the rear seat of minivans with sliding doors

This study investigates the motion strategies performed by elderly and young passengers while entering and exiting the rear seat of minivans with sliding doors. A minivan mock-up was constructed with four adjustable parameters to represent nine different conditions of vehicle geometry. Ten elderly male participants (66.8 ± 3.8 years old) and ten young male participants (31.5 ± 6.6 years old) were recruited. Each of them entered and exited the minivan mock-up for five times under each condition, and the motion data were acquired by the optical motion capture system. Based on the criteria derived from previous studies, all motions were automatically categorized into seven ingress motion strategies and seven egress motion strategies. Further, the differences among motion strategies are discussed in terms of vehicle factors and passenger factors, which provide clues for future studies.

Forward dynamics simulation of human figures on assistive devices using geometric skin deformation model

We present a forward dynamics (FD) simulation technique for human figures when they are supported by assistive devices. By incorporating a geometric skin deformation model, called linear blend skinning (skinning), into rigid-body skeleton dynamics, we can model a time-varying geometry of body surface plausibly and efficiently. Based on the skinning model, we also derive a Jacobian (a linear mapping) that maps contact forces exerted on the skin to joint torques, which is the main technical contribution of this paper. This algorithm allows us to efficiently simulate dynamics of human body that interacts with assistive devices. Experimental results showed that the proposed approach can generate plausible motions and can estimate pressure distribution that is roughly comparable to the tactile sensor data.

Multiple regression based imputation for individualizing template human model from a small number of measured dimensions

Individual human models are usually created by direct 3D scanning or deforming a template model according to the measured dimensions. In this paper, we propose a method to estimate all the necessary dimensions (full set) for the human model individualization from a small number of measured dimensions (subset) and human dimension database. For this purpose, we solved multiple regression equation from the dimension database given full set dimensions as the objective variable and subset dimensions as the explanatory variables. Thus, the full set dimensions are obtained by simply multiplying the subset dimensions to the coefficient matrix of the regression equation. We verified the accuracy of our method by imputing hand, foot, and whole body dimensions from their dimension database. The leave-one-out cross validation is employed in this evaluation. The mean absolute errors (MAE) between the measured and the estimated dimensions computed from 4 dimensions (hand length, breadth, middle finger breadth at proximal, and middle finger depth at proximal) in the hand, 2 dimensions (foot length, breadth, and lateral malleolus height) in the foot, and 1 dimension (height) and weight in the whole body are computed. The average MAE of non-measured dimensions were 4.58% in the hand, 4.42% in the foot, and 3.54% in the whole body, while that of measured dimensions were 0.00%.Individual human models are usually created by direct 3D scanning or deforming a template model according to the measured dimensions. In this paper, we propose a method to estimate all the necessary dimensions (full set) for the human model individualization from a small number of measured dimensions (subset) and human dimension database. For this purpose, we solved multiple regression equation from the dimension database given full set dimensions as the objective variable and subset dimensions as the explanatory variables. Thus, the full set dimensions are obtained by simply multiplying the subset dimensions to the coefficient matrix of the regression equation. We verified the accuracy of our method by imputing hand, foot, and whole body dimensions from their dimension database. The leave-one-out cross validation is employed in this evaluation. The mean absolute errors (MAE) between the measured and the estimated dimensions computed from 4 dimensions (hand length, breadth, middle finger breadth at proximal, and middle finger depth at proximal) in the hand, 2 dimensions (foot length, breadth, and lateral malleolus height) in the foot, and 1 dimension (height) and weight in the whole body are computed. The average MAE of non-measured dimensions were 4.58% in the hand, 4.42% in the foot, and 3.54% in the whole body, while that of measured dimensions were 0.00%.

Object-dependent estimation of grasp posture and contact region of hand based on cluster analysis

This paper presents a data-driven framework for estimating grasp posture and contact region between hand and object by using clustered lower dimensional grasp features. Our framework extracts features of the grasp posture in order to estimate the contact region. Redundant dimension of the hand posture is removed depending on task, which we regard typical for the object. For this purpose, we used mixture principal component analysis (MPCA). This method enables to estimate clusters that can approximate grasp postures by using lower dimensional features. Estimation results of the contact region was obtained by using clustered lower dimensional features of MPCA and object features. These results show that our clusters by MPCA have stronger correlation with the contact region than clustered lower dimensional features obtained by PCA and k-means. Finally, estimation results of the grasp posture and the contact region for three objects are demonstrated. One of the result was compared with grasp posture synthesized by another method presented in the previous research. This comparison revealed that our framework was able to synthesize human-like natural grasp posture than the previous one.