Evolutionary Design Method of Morphology and Intelligence in Autonomous Mechanical System. 2nd Report, Generality and Inner-Structural Characteristics of Morphology/Intelligence.

Abstract

This paper deals with function generality and inner-structural characteristics of morphology/intelligence in autonomous mechanical (robotic) system which is generated using evolutionary design method. In the realistic virtual world including physics such as gravity, collision and friction, the robotic systems generated are found to have various morphologic feature and intelligence. In performing a given task, they show a various morphologies together with the corresponding interesting strategy. At first, the effect of noise addition to the initial condition on the task performance is discussed, and then we analyze the complexity of the robotic morphology and "dimension consistency" in the intelligence described by the decision tree expressing the inner-structures of the robotic system. It has found that 1) a small noise addition has improved markedly the task-performance rate, and also the robustness of the robotic system, and 2) the physical-dimension-consistency can explain the inner-structural character of the robotic intelligence that realized the better task performance.