We are presenting in this work a method to calculate collision free paths, for redundant and non redundant robots, through an adaptation of the Messy Genetic Algorithm with a fitness function weakly defined. The adaptation consists in replacing the two crossing operators (cut and splice) traditionally used by a mechanism similar to that one used in the simple genetic algorithm. Nevertheless, the mechanism presented in this work was designed to work with variable length strings. The main advantages of this method are: even though the fitness function is weakly defined good solutions can be obtained; it does not need a previous discretization of the work space; and it works directly within such space without needing any transformation as in the C-space method. In this work, the fitness function is defined as a linear combination of values which are easily calculated.