First Translation Into a GA

Figure 9.2: Applying the GA Schema onto the Scenario

The interesting question is, whether -and if yes: how- one can apply the concept of a Genetic Algorithm (GA) onto the above experimental scenario. The basic ideas of a GA can can be founbd in chapter about GAs. In figure 9.2 one can see a sketchy argument how to map the structure of a GA onto the experimental scenario. The basic idea is as follows:

1. As the interesting population $P$ a string of atomic motor movements has to be identified where every atomic movement (L, R, O, U, N) can be represented as a binary pattern. The population $P$ can also be interpreted as the 'content of a simple memory' which can grow and change.
2. The fitness $F$ will be generated by a comparison between those perceived bitmaps which represent the movements on the 'own' body and those perceived bitmaps which represent the movement of a 'different' body. As higher the similarities are as higher should be the fitness with regard to imitate the movements of the other body.
3. The perceived bitmaps are caused by the real movements of bodies on a sens floor. Touching on a sensor in the floor is causing a '1', otherwise the floor shows a '0'. It is assumed that the whole floor is presented to the perceptive system of the learner as one big bitmap having the 'directions' 'O', 'U', L', 'R'.
4. In a simplified fashion we assume for the beginning that only one body is present on the floor, either the body of the learner or another body.
5. In the beginning we assume further that every body on the floor is acting for a finite amount of cycles between 3 and n.
6. The movements of the learner body are triggered by motor events which are corresponding to the atomic motor movements.

Remark to fitness: As can be seen from the general chapter about GAs is the fitness a dynamic value describing the relationship between the behavior of a system and its environment with regard to a certain criterion. In the case of imitation learning the criterion is the similarity between the given values from the environment compared to the selfproduced values. As higher this similarity is as higher is the success of the imitation. Therefore one can use this similarity as starting point for the compuation of the fitness of those elements of the system which are causing the selfproduced behavior. In the above case is this given by those motor patterns which are triggering the movements.