A multi-agent approach of the multi-sensor fusion

Describes the study and the realization of an evolutive perception system for a mobile robot which evolves in an unstructured environment. The integration of multisensor fusion, which takes a lot of data and the importance of the algorithms into account, leads one to consider a functional architecture which is adaptable and modular. The perception system has to be able to decide its own configuration (active sensors and processing), according to the environment, the mission, the dynamic ability of the robot and the available sensors. The implementation is based on a blackboard model. The communication and cooperation between agents permit a large amount of interactivity between the action, perception and decision. Finally, the implementation and an application are described.<<ETX>>