Alexandru Horia Brie and Philippe Morignot
This paper describes a genetic planning system, i.e., a program capable of solving planning problems using evolutionary techniques. As opposed to other approaches in Genetic Planning, we use a variable length chromosomes model in addition to a complex fitness function and several enhancements of the Simple Genetic Algorithm (Holland 1975), such as multipopulations, population reset, weak memetism, tournament selection and elitist genetic operators. Our genetic planner is tested on standard planning domains and problems (described in PDDL), is used for parameter and performance analysis, and is compared to previous work. Results show efficiency in memory management and greater solving power than its predecessors'.