Architectures de contrôle comportementales et réactives pour la coopération d'un groupe de robots mobiles

Abstract

The control of a highly dynamic multi-robots system where evolves a large number of reactive autonomous entities constitutes a promising scientific and technological challenge. Indeed, this requires to use robotics entities as elementary as possible, but also requires in term of control, to move more away from centralized and cognitive designs. The approach consists in focusing the control design more on the elementary individual constituting the multi-robots system while taking into account its different local interactions with other robotics entities, which are supposed to cooperate between them. Mastered mass effects can be gotten thus and will permit to enhance at the same time the speed, flexibility and the robustness of the executed complex tasks. The works of research presented in this thesis leave from the principle of a Bottom-Up architectures of control in order to break the inherent complexity of the multi-robots systems. More specifically, we propose a Hierarchical Action Selection Process named HASP which permits at the scale of the robot to coordinate the activity of a set of elementary primitives (behaviors) in a hierarchical and flexible manner, and at the scale of the group of robots, allows a coordination promoting global goals. The performances of the HASP were improved after via the addition of an appropriate mechanism of fusion of actions leading thus to the HHASP (Hybrid-HASP). The formalisms of the genetic algorithms have been used afterwards to propose a methodology, which allow to obtain preponderant parameters for the working of the HHASP. The validation of the results was made through experimentations with mini-robots ALICE and more intensively on statistical studies achieved on a big number of data gotten thanks to MiRoCo (Mini-Robotique Collective) simulator. This simulator have been designed and developed in the framework of our thesis works in order to simulate accurately and rigorously multi-robots systems, which have a strong dynamic of interaction.