Dynamic configuration of mobile robot perceptual system

This paper describes SEPIA, a system that integrates reactive behavior in a control architecture allowing the configuration of the perceptual system according to execution context, robot objectives and sensor performances. SEPIA is based on multi-agent system concepts where the perception, command and planning functionalities are modeled by agents. After the presentation of the architecture based on a blackboard idea, the authors define the agent structure and model. This model is necessary for an efficient collaboration between agents using the valuation of the performances according to its ability and the context. The selective references are based on result quality and response time t. The dynamic reconfiguration is made possible by bids mechanism associated with selective rules on reference criteria. Finally the implementation of SEPIA is described and the results of experimentation on a real mobile robot are discussed