Evolutionary Robotics for Self-Organising Behaviours

As mentioned in the previous chapter, there is a fundamental problem—referred to as the design problem—that arises in the development of selforganising behaviours for a group of robots. This problem consists in defining the appropriate individual rules that will lead to a certain global pattern. In Section 4.1, we analyse in detail which are the difficulties that may be encountered in developing a control system for a group of robots. Such difficulties can be bypassed resorting to evolutionary robotics as a design technique, as described in Section 4.2.

Evolutionary robotics is an automatic technique for generating solutions for a particular robotic task, based on artificial evolution (Fogel et al., 1966; Holland, 1975; Schwefel, 1981; Goldberg, 1989). It is inspired by natural evolution, which predicates the “survival of the fittest”: the individual that best adapts to its environment has more chances to reproduce and to pass its genetic material to the subsequent generations. In this way, the species evolves toward better and better individuals. The same idea is exploited in the artificial counterpart, in which a population of individuals is evolved for many generations. Each individual, characterised by its genotype, represents a solution for a given task. Its fitness—i.e., the quality of the solution to the task-is automatically evaluated in each generation. The “fittest” individuals are allowed to “reproduce” by generating copies of their genotypes. The latter are modified using genetic operators, such as crossover (sexual reproduction) or mutation (asexual reproduction). In this way, offspring are generated that undergo the same evaluation process, until a valid solution is found. Figure 4.1 gives a schematic description of this process.