DIGITAL HUMANS AND HUMANOIDS

by Dr. Masaaki Mochimaru

Digital Human Research Center,

National Institute of Advanced Industrial Science and Technology, Japan

Humans, the weakest link

The human is an essential element of man-machine systems. In fact, the human is the most important component of the system, but at the same time it is the least understood element. Consider a golf player, as an example. A human and a golf club form a system intended to strike a golf ball as far and as accurately as possible. A golf club is designed with advanced technologies including a titanium head, a carbon shaft, and a finite element analysis of the prototype. Yet, the player's movements are not understood completely. Therefore, the human is the weakest link of the system.

The purpose of the Digital Human Research Center (DHRC) is to study and understand human functions. This is achieved by conducting computer simulations and utilizing the computerized humans in industrial applications. Professor Takeo Kanade, a robotic researcher and the director of the Robotic Research Institute at Carnegie Mellon University, launched the DHRC in 2001. Then the National Institute of Advanced Industrial Science and Technology (AIST) invited him to serve as the director of a new laboratory. He proposed the concept of Digital Human Research and opened a small lab with six researchers. Since then, the DHRC has expanded and currently employs approximately 40 people, including 14 permanent researchers, seven postdoctoral fellows, and six technical staff.

The lab focuses on three functional aspects of humans. The physio-anatomical aspect studies the living body, the motion-mechanical aspect studies the robotic mechanisms of the human body, and the psycho-cognitive aspect concentrates on the human systems used to feel and decide. In order to develop a computational human model that can accommodate all three aspects, modeling and simulation as well as observation and presentation are required. The research covers a broad range of fields, so the lab includes members with a variety of backgrounds, including biomechanics, robotics, virtual reality, computer vision, computer graphics, informatics, physical anthropology, clothing engineering and medicine. The four research groups are human centered design, human behavior understanding, humanoid interaction, and digital human modeling. All the groups utilize a motion capture system for their research. This paper presents three topics related to the motion capture. Additional information on this research can be found at http://www.dh.aist.go.jp

Genuine digital manikin project

The objective of this project is to develop a method for assessing the interaction between a car and humans using whole body digital humans. A digital manikin is modeled using articulated links. Anthropometric databases are used to calculate the size of each link, and several models can be generated to represent the variation in the human population. The posture of a model can be adjusted by stipulating the position of the end effectors, the hands and feet, using inverse kinematics. The reach envelope and visual field are simulated for product assessment. The advanced technologies employed by digital manikins have been refined in the United States and Germany, and several commercial products are currently available. However, several important problems remain, including the anthropometric accuracy of the functional dimensions, automatic motion generation, and the perception of discomfort. DHRC launched a consortium in 2003 to address these problems. Eight universities, three institutes, five automotive companies, one housing equipment company, and seven software developers participated. Additional information on this consortium is available at http://www.digital-human.jp

Two motion-capturing projects are currently in progress: modeling the shoulder joint and modeling the alternative motions for entering a car. See Figures 1 and 2 on page 1. The aim of the first project is to develop an anthropometrically accurate shoulder model. The center of the shoulder joint is defined as the center of rotation for the proximal part of the upper-arm segment. A new method was developed to estimate the shoulder joint center that uses surface markers pasted on the upper arm. In this method, a transformation matrix between the joint center and the marker coordinate system was calibrated by measuring a specific predefined shoulder motion for each subject. This method is depicted in the upper portion of Figure 2. This transformation matrix was assumed to be constant throughout the motion. The joint center was also assumed to be fixed in the global coordinate system over a very short period of time. Based on these assumptions, the transformation matrix was optimized using the measured marker data during the predefined shoulder motion. The relationship between the estimated joint center and humerus was also examined using MR images of a subject equipped with the marker set. The estimated joint center was located on the surface of the humerus head, as shown in the lower portion of Figure 2.

The second research topic involves modeling alternative motions. Automotive companies used car mock-ups to determine the motions required to enter a car. Figure 1 shows two examples. These two different subjects employed drastically different motion sequences on the same mock-up even though they were nearly the same height. One subject entered with his foot, as shown in the top left and middle schematics in Figure 1, and the other subject entered with his head, as shown in the top right and bottom schematics in Figure 1. The objective is to quantify the differences between the two motions, to classify the alternatives, and to generate additional possibilities.

Digital hand project

The primary objective of the digital hand project is similar to that of the digital manikin. The digital manikin project focuses on the entire body structure, shape, motion, and loads, while the digital hand project concentrates on the detailed hand structure, shape, and motion. Moreover, the reaction forces, deformation, and finger friction will be simulated and the slip sensation, pressure sensation and manipulation perception will be estimated. The ultimate goal is to create an integrated functional hand model that incorporates all three human functional aspects: physio-anatomy, motion-mechanics and psycho-cognition. For this study, the hand was modeled with twenty articulated link segments containing forty degrees of freedom. Eight Vicon cameras (Vcam 624, 200 Hz) and two lipstick-sized infrared Vicon cameras (SVcam 624, 200 Hz) were used to capture the hand's motion. The cameras were focused on a 300 cubic millimeter measurement area. The two infrared cameras were employed for the purpose of avoiding light emitted from cameras from entering other cameras. Twenty-five 5mm surface markers were attached to the hand, as shown in Figure 3. Although the hand is small, it boasts a high number of degrees of freedom. Theoretically, it would be optimal to attach three surface markers to each segment. However, this is not practical due to the potential for occlusion and mislabeling. Therefore, the hand's forty degrees of freedom were measured using only twenty-five markers. The individual hand model's link parameters, including the link length, joint axis, and joint center, were calibrated by performing a specific predetermined hand motion. With this method, errors in the fingertip locations between the hand model and actual hand were less than 2.0mm. This accuracy was deemed acceptable for the tasks performed for this project. Figure 4 shows examples of the reconstructed postures obtained by this method.

Human and Humanoid

DHRC research topics included human-like robots, called humanoids, that can function in the real world. The final goal of the humanoid project is to develop an autonomous robot that can recognize and understand an environment, generate movement trajectories and joint motions, and control both dynamic and static motions. This section provides an overview of one of the current humanoid projects. The project involves comparing the walk of a human and a humanoid. The purpose of this study is to understand how a real human controls his center of gravity (COG) and how this information can be applied to a humanoid control to improve humanoid bipedal walking. First, the gait of humans and the humanoid were measured and compared. The experimental set-up included a 10meter by 3meter walking path equipped with eight Vicon cameras (Vcam 624, 200 Hz) mounted on the ceiling and two infrared Vicon cameras (Vicon 624, 200 Hz) installed on the ground. Five force platforms (AMTI, 0.6m x 0.4m, 1000 Hz) were placed along the center of the walkway. Two of the platforms (1.2m long by 0.4m) were placed on the left side and three platforms (1.8m long by 0.4m) were placed on the right side. The motions of the entire body were captured with the marker set shown in Figure 5. Next, the motion data were exported in a DIFF file format that was proposed by the Clinical Gait Analysis Forum in Japan. The joint angles, the zero moment point (ZMP) trajectory on the ground, and the center of gravity for the entire body were calculated by gait analysis software produced by the Clinical Gait Analysis Forum in Japan. The humanoid's ZMP trajectory, which was calculated as the zero moment point of vertical and horizontal ground reaction forces, was found to be similar to that of the human. For dynamic balancing, the ZMP must be located in the foot. Because the humanoid's body size and step width are similar to a human's, it is reasonable for them to have similar ZMP trajectories as well. However, the major difference between the human and the humanoid was observed in the COG trajectory. The humanoid's horizontal COG displacement is much larger than that of the human, whereas the human's vertical COG displacement is much larger than that of the humanoid. The ZMP formula includes the horizontal COG acceleration in the numerator and the vertical COG acceleration in the denominator. The humanoid avoids vertical COG displacement to save energy, but it matches the ZMP trajectory by imposing a large horizontal COG displacement. In contrast, the human generates its ZMP trajectory with a large vertical COG displacement while minimizing lateral COG movement. The humanoid's motion control sequence was changed to approximate a vertical sign wave COG displacement. As a result, horizontal body sway of the humanoid was reduced by 80% and the maximum gait speed was increased by 50%. Therefore, examining the human gait helped improve a humanoid's bipedal walking.

References

Kouchi, M., M. Mochimaru and M. Higuchi (2004). A validation method for digital human anthropometry: towards the standardization of validation and verification. SAE Digital Human Modeling Conference (DHM2004), Detroit, US.

Miyata, N., M. Kouchi, T. Kurihara, and M. Mochimaru (2004). Modeling of Human Hand Link Structure from Optical Motion Capture Data. Proc. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS2004), Sendai, Japan.

Ryu, J. H., N. Miyata, M. Kouchi, M. Mochimaru and K. H. Lee (2003). Analysis of skin movements with respect to bone motions using MR images. IJCC Workshop ‘03: Digital Engineering, Jeju Island, S. Korea.

Kagami, S., M. Mochimaru, Y. Ehara, N. Miyata, K. Nishiwaki, H. Inoue and T. Kanade (2003). Measurement and Comparison of Humanoid H7 Walking with Human Being. IEEE International Conference on Humanoid Robots (Humanoids2003), Munich, Germany.