仮想進化の試み : 跳躍動作の進化シミュレーション(2部生体の特性)

概要

論文の詳細を見る
In recent years, the mechanism of organic evolution has been investigated by computer simulation in the field called artificial life. The creatures simulated in these studies are mainly lower and abstract ones, and their body dynamics are not considered. However, biomechanical factors such as muscle tension certainly play important roles in the evolution of animals. That is, motions are reasonably adapted to the mechanical constraints of the body system, whereas a repetitive motion modifies body construction adapting to its motion. Clarifying the mechanism of interactive adaptation between body shape and repetitive motions is helpful for studying evolutional processes in animals. This study focuses on the evolutional adaptation observed in animals and the acquisition process of animal hopping motions, which are considered to be strongly influenced by biomechanical factors, was reproduced using the techniques of computer simulation and a neuro-musculo-skeletal model. A body model was constructed of seven rigid links in the sagittal plane, in which the right and left legs were made up into one leg. These links were driven by twelve muscles. The nervous system was modeled into a rhythm pattern generator consisting of twelve neural oscillators. Hopping motions were generated through mutual entrainment between the neural oscillation and the body dynamics of the musculo-skeletal system. The evolutional algorithms that change body motions and body shape consist of acquired processes corresponding to neural learning and inherent processes using genetic algorithms. Firstly, a group of individuals was generated according to genes. In the acquired processes, neural connectivity was modified by evaluating the adaptability of the synthesized hopping motions for each individual. In the inherent processes, individuals whose value of fitness was small and those who fell down were weeded out. In addition, genetic operations such as mutation produced various body shapes. In the initial stage of evolution, continuous and high-speed hopping motions were not synthesized. First, the locomotive-speed gap between objective and actuality, specific power, and moment of inertia of each leg were defined as the criteria for motions and fitness for genetic algorithms. As selection and multiplication were repeated, high-speed quadruped hopping motions were obtained. Second, minimizing the ground reaction force of the foreleg was added to the criteria and fitness mentioned above. In this case, bipedal hopping motions were obtained, and the body shape became kangaroo-like. From these results, it is thought that adaptation to high-speed locomotion and facility in motion control are important factors in the evolution of hopping motions and body shape. Virtual evolution, such as this study, not only provides a novel methodology for research on animal evolution, but also can be applied to various fields such as mechanical engineering.
バイオメカニズム学会の論文
1998-11-25

仮想進化の試み : 跳躍動作の進化シミュレーション(2部生体の特性)

スポンサーリンク

概要

著者

関連論文

スポンサーリンク

仮想進化の試み : 跳躍動作の進化シミュレーション(2部 生体の特性)

スポンサーリンク

概要

著者

関連論文

スポンサーリンク

仮想進化の試み : 跳躍動作の進化シミュレーション(2部生体の特性)