利用基因演算法自動產生機器蛇攀階步態

Abstract

本研究提出一套使關節串接型機器蛇朝向攀階目標位置移動之一般化架構，並運用基因演算法（Genetic Algorithms）自動產生攀階步態。由於目前機器蛇攀階時缺乏系統化的方法以產生攀階步態。本研究依此架構發展基因演算法的適應性函數作為攀階過程的評分標準。此標準分為六階段，以各階段提供期望中的空間絕對座標做為評估標準，建立一般化規則。此規則可根據機器蛇每節長度與節數、樓梯高度以及空間位置的關係，推算各階段中各節的座標位置。本研究結合Matlab基因演算工具與Webots移動式機器人模擬軟體，配合上述六階段的適應性函數，產生一系列的攀階角度，即所謂的步態，並將所得之攀階步態應用於真實系統中。實測結果顯示，在機器蛇節數為12、樓梯高度為16公分之情況下，機器蛇執行初始攀階步態後，各節位置與目標位置的平均差異為7公分，在同樣執行個體數為30、經歷30個演化世代之步態後，平均差異降至4公分。據此，可證明本研究所提出之一般化適應性函數評分標準，可達到自動產生較佳的攀階步態之目的。增加演化個體或演化世代，一般可使演化的步態更接近期望之目標位置，但相對地演化的時間亦將拉長。使用者可視實際需求，在精準度與演化時間之間做出取捨。

In the study, we propose a generalized structure for articulated snake robots to move towards the aim position when ascending the stairs. In addition, we exploit Genetic Algorithms to automatically generate stair-ascending gaits for snake robots. Since there is no systematic approach for planning stair-ascending gaits, we base on the proposed structure to generate such gaits and to establish a generalized GA fitness measure for evaluating the process of climbing stairs. The fitness measure falls into 6 stages, each of which specifies the absolute positions for each module of the snake robot as the evaluation standard. We conclude the standard and develop a generalized fitness measure. According to the sizes of each module, the total number of modules of the snake robot, the height of the stair, and the spatial relationships between them, we can infer all the absolute positions for each module of the snake robot in each stage. We further integrate Matlab Optimtool GA and Webots, a mobile robot simulation platform, along with the 6-stage fitness measure, to generate a series of angles for each module of the snake robot to accomplish the task of ascending stairs. Following that, we apply the series of angles, or the gait, to the snake robot in the real, physical world. One real-world result shows the gait for a 12-module snake robot with 16 cm stair is improved from an average deviation of 7 cm for each module between the final position of the snake robot and our aim to an average deviation of 4 cm for each module within the evolution of only 30 individuals for 30 generations, meaning that the integration of Matlab Optimtool GA and Webots along with our 6-stage fitness evaluation measure works for automatically generating the stair-climbing gait for snake robot to better move towards its aim position. Generally speaking, the resultant gait can be improved, making the snake robot reach closer to its aim position, by increasing the individuals or the generations in GA evolution, yet it may relatively take more time for GA to produce the resultant gait. Therefore, users can strike a balance between the final position of the snake robot and the time for GA evolution according to their needs.