1.1 Motivation and Purpose
At present, robots are found in many different applications not limited to industrial arms, exoskeleton legs, and building security [1]. In the future, robotic applications are expected to extend to the development of full body humanoid robots for purposes such as assistive care, disaster recovery, ballroom dancing partners, and so forth [2].
However, current robots do not accurately mimic the motions and postures of real humans due to different dynamic parameters. In developing humanoid robots, the problem of maintaining balance during bipedal motion is particularly challenging [3].
Furthermore, existing humanoid robot designs are based on separate models of links in the human body (e.g., the legs, trunk, arms, hands, and so on). In creating a humanoid robot, it is necessary to integrate these various models, resulting in motion which tends to be unnatural or non-fluent. Consequently, developing a more realistic model of whole complete human body motion is urgently required.
This study proposes a method for determining the dynamic parameters of human motion using a 3D whole human body model. In the proposed approach, a two-phase optimization process is used to solve the redundant inverse kinematic problem, such that the human motion can be described directly in terms of the joint angles. In addition, the dynamic parameters of the whole body are calculated by integrating the dynamic data of the constitutive links. Furthermore, a new approach based on the screw method is presented for tracking the ZMP of the human body during motion.
The validity of the proposed approach is demonstrated by evaluating whole body dynamics over the course of a 25-second sequence of continuous moves and postures
performed by a professional martial arts practitioner. Finally, reaction forces acting on the human body are analyzed according to a screw formed by the human body and feet positions.
1.2 Literature Review
Previously, Dasgupta & Nakamura [4] successfully used motion capture data to generate natural motions for humanoid robots. Humanoid robot motions are controlled using joint angles instead of captured motion data. Thus, inverse kinematics is used to solve joint angles from the body motion data.
Pollard et al. [5] realized a humanoid robot capable of imitating the motion of the human upper body by simply converting the joint angles observed in real human motion. However, the motion of the lower body was not considered. Ruchanurucks et al. [6] proposed a more sophisticated method for optimizing the upper body motion of a humanoid robot, based not only on the joint angles but also the basic characteristics of the corresponding motion. Furthermore, in constructing the body model, the inherent constraints of humanoid robot structures (e.g., a limited range of motion or a reduced degree of freedom (DOF) of the joints) were taken into account.
However, the problem of maintaining whole body balance during motion was not considered. Nakaoka et al. [7] proposed a method for determining the leg motion primitives of a humanoid robot by solving an inverse kinematics problem based on the motion data captured during a human performing a sequence of dance moves. In the proposed method, the lower limb joint angles were modified in accordance with the velocity constraints of the joint motors, while the leg motions were modified in accordance with the desired ZMP trajectory. However, in order to maintain balance,
the robot was obliged to perform dance motion at a speed equal to around just one half that of the actual human motion.
Achieving body balance involves the dynamics of the entire body. However, humans and humanoid robots have multiple extremities, and thus the link dynamic parameters are highly complex. Moreover, adjusting the links one by one, so as to maintain balance, poses a significant challenge. Burnfield [8] found that a short period of unbalance occurs during normal human movement as the weight of the body is transferred from the rear leg to the forward leg. Nonetheless, given an expression for the dynamics of the whole body, the balance problem can be rapidly solved, thereby paving the way for the development of robots with a more humanlike motion.
Firnami & Park [9] proposed a theoretical method for analyzing the whole body based on classic mechanical principles. The proposed method quantifies the imbalance state by means of the distance between the center of pressure (CoP) and the ZMP. The obtained results confirm that humans experience a forward fall at the end of the single-stance phase of bipedal motion. However, the model is skeleton-shaped rather than truly human skeleton-shaped. Moreover, the body segments are formed by means of node points [9], therefore, inaccuracies may occur in estimating the true geometric parameters (e.g., centroid, moment of inertia (MOI), and principal axes) and determining the shape of the support area. Consequently, the dynamic parameters and unbalance calculations may also be prone to error.
Ren et al. [10] proposed a study to deal with underdetermined problems of the external contact loads and represented the load sharing in double support phase with mathematical functions. The method provided satisfying results, but it was specific to the given task.
1.3 Outline
The remainder of this paper is organized as follows. In chapter 2, human model profiles are introduced to perform human motions and calculate geometric parameters of each body segment. Chapter 3 describes the two-phase optimization approach used to convert captured body motion data into equivalent joint angle motions. In chapter 4, the dynamic parameters of the whole body are derived by integrating the dynamic data of all the links. In chapter 5, the screw theory is used to determine the ZMP is introduced, and the simulation result is presented. In chapter 6, foot reactions are derived using the screw theory and analyzed according to optimization results. Finally, chapter 7 provides some brief concluding remarks and indicates the intended direction of future research.