Auto-generated Control Program in Mobile Robot using Grammatical Evolution

Abstract

Complexity of robot control program is become a problem as variety of robot used today is dramatically increased due to demand of automation. Several studies has been made using GE to solve robot tasks but application in actual environment is still lack of proof. In this research, auto-generated control program using Grammatical Evolution (GE) is proposed to resolve the burden of constructing manually adjusted control program that consider both physical condition of robot and input from sensors. Grammatical Evolution (GE) is a genetic programming that map genotype to phenotype using predetermined grammar. As GE is subset to Genetic Programming which produce heuristics solution, the grammar-based algorithm enable the search range evolve in a set of rules. This search strategy can be applied to build the structure of control program which consists of multiple sets of program blocks with adjustable parameter. This system is evaluated using Santa Fe Trail problem with open world configuration. The results show the usability of the algorithm by manipulating crossover and mutation rates. Another important factors that effect the performance of the algorithm is the grammar that can govern the evolution of the control program. The proposed method able to generate control automatically with sets of rules determined using grammar.