Model-based Control Architecture for Attentive Robots in Rescue Scenarios

In this paper we present a control architecture for an autonomous rescue robot specialized in victim finding in an unknown and unstructured environment. The reference domain for rescue robots is the rescue-world arenas purposefully arranged for the Robocup competitions. The main task of a rescue mobile robot is to explore the environment and report to the rescue-operators the map of visited areas annotated with its finding. In this context all the attentional activities play a major role in decision processes: salient elements in the environment yield utilities and objectives. A model-based executive controller is proposed to coordinate, integrate, and monitor the distributed decisions and initiatives emerging from the modules involved in the control loop. We show how this architecture integrates the reactive model-based control of a rescue mission, with an attentive perceptual activity processing the sensor and visual stimuli. The architecture has been implemented and tested in realworld experiments by comparing the performances of metric exploration and attentive exploration. The results obtained demonstrate that the attentive behavior significantly focus the exploration time in salient areas enhancing the overall victim finding effectiveness.