Improving Naturalness of Locomotion of Many-Muscle Humanoids

Abstract

For many decades, researchers have worked on the simulation of biped locomotion such as human walking. As simulation models have evolved, the simulation using a musculoskeletal model has also become possible [Lee et al. 2014]. Since the number of muscles of models they use is greater than the number of DoF of the models, the optimization problem is undetermined. In order to solve this problem, they use the 2-norm of muscle activations as an objective of optimization and minimize it. However, to obtain more realistic simulation results, real mechanisms of human movement should be applied. In spite of development of the humanoid locomotion simulation, it is still not natural enough due to the lack of knowledge in the mechanisms of human movement. Discussions about this topic have been made, and many people believe that human movement tries to minimize one of the followings; muscle activation, derivatives of muscle activation, joint torque, derivatives of joint torque, metabolic energy expenditure or some combination of these. In this study, we did experiments for each minimization case and compared the results of kinematic data and energy consumption.