A hybrid architecture for robot navigation

In the context of the approach to intelligent autonomous systems based on the subsumption architectural concept, the authors describe a hybrid model of the navigation skill, to be considered as one of the many skills or behaviours that allow an autonomous agent to survive in an unknown/hostile environment. The hybrid navigation behaviour consists of three main functions that operate in parallel on the same set of input/output data: (i) WRA (wild rover algorithm), (ii) SON (self-organized navigator), (iii) SEA (symbolic environment analysis). The term "hybrid" here refers to the cooperation between a logics-based representation formalism and a neural model. Starting from rough sensorial data given by WRA, the knowledge about the explored environment of a mobile robot can be incrementally organized by means the self-organizing maps (SON) and the set of heuristic rules (SEA).