The organizators of the contest appreaciated the quality of all demonstrations which made the contest of high interest for the evolutionary computing community present at the conference. Many people were very enthousiast seeing the contest and promised they would engage the next Khepera contest.
The problem consist mainly in exploring a previously unknown environment
while detecting, registering and recognizing light sources which may dynamically
be turned on and off. At the end of each round a report is requested from
the robot. Therein, an agent-based approach was used and a self-organizing
feature map was applied in order to refine some of the behavior generating
control modules.
Appraisal is a theory that has been developed in study of emotion based
upon the capacity to evaluate a stimulus following some dimensions, without
having to recognize it. The neurobiologist J. LeDoux (1996) (That resumes
partly ideas of MacLean (1970)), and A. Damasio (1995) suppose an essential
role of emotion in the phylogenetic of brain. It is suggested that appraisal
is an adaptative mechanism, that allows to react even in the absence
of knowledge and this of rapid manner. Resuming this idea of appraisal,
three controllers (based upon neuronal network) have been designed. The
first is based on appraisal model of Scherer (1993,1997). The second is
a hierarchical decision-making system. The third regroups the two firsts.
These three controllers are tested following two tasks ( 1) avoidance of
obstacle, 2) tracking a luminous point with obstacle avoidance) and in
two types of environment (static and dynamic). It is shown that a
controller resting solely on the appraisal can have a success score equal
to a controller resting on a decision-making system, and that the addition
of an appraisal part allows a better rate of success in dynamic environments.
At the end, perspectives of appraisal application are pointed, notably
in learning.
Artificial neurogenesis seems to be a promising way to evolve neural networks. It doesn't use a direct encoding of networks on the chromosomes. Chromosomes contains instructions to grow a network from an initial seed. This growing process is called morphogenesis or neurogenesis. The neurogenesis method developed is strongly inspired by the natural protein regulation process. Some of the proteins corerspond to particular functions implying specific behavior for the for the cell, such as splitting or connecting behavior. This method is applied to evolve neural controllers for the Khepera robots. The system is made up of three parts interconnected in an evolutionary loop. The first part is a genetic algorithm, the second part is the morphogenesis process and the third part is the simulation and evalution of the behavior of the robot. Evolution is done in simulation and results are finally transfered to the real robot.
The control of a robot and automatic obstacle avoidance is considered for a low-level system. The aim of this project is to show how a simple set of fuzzy rules can be sufficient to obtain a good behavior of a robot into an unknown environment comprising fixed obstacles.