The methods of generating initial populations in tree-based genetic programming, and applying them to create robot’s movement program within bounded area

Abstract

In this research, we aim to study the methods of generating initial populations in tree-based genetic programming and using them in generating a program for robot movement. This study includes several methods, which are, Grow method, Full method, Ramped Half and Half, Uniform method, Probabilistic Tree-Creation 1 and 2, We applied these methods in generating programs for a multiple path tracker robot and we simulated these methods on artificial ants’ problems, which are Santa Fe trail problem, John Muir trail problem, and Los Altos trail problem, and adding modified paths to this research in order to study the effect of changing multiple factors such as increasing the number of turns in paths, increasing the number of subpaths, and increasing path length.  We found in this study that the best method was the uniform method, because of the size of the correct programs was 66% less than the size of the correct programs generated other methods, then PTC2 at 51%, Grow method at 47% and PTC1 at 45%, and we found that the increment in program size had a negative impact on the program’s fitness, the number of Evaluations, and on the percentage of correct programs.