Apprentissage et adaptation pour des ensembles de robots cooperants

When a group of robots manoeuvres in an unfriendly environment, teleoperation is not always possible, and traditional methods of planning are generally time consuming. Their learning and adaptation capabilities are constantly increasing. Our purpose is to make robots learn a complex collective task without the need for human intervention. This paper describes two methods based on evolutionist algorithms and simulated annealing. Both ways have been implemented firstly by simulation and secondly on a real platform composed of 4 mobile robots presented in the third part. The fourth part describes the actual results.