NAVNEX: an hybrid system which learns navigation situations in real time

The paper introduces NAVNEX (Navigation and Neural Explorer), a hybrid system, neural and symbolic/procedural, well suited for real time extraction of navigation situations during incremental explorations of unknown environments performed by an autonomous mobile robot. Real time performances are obtained through the parallelization of the activities necessary to extract relevant information: reactive planning and navigation, environment reconstruction via self-organization, symbolic (topological) analysis of reconstructed maps. The neural part of the system is based on two self-organizing neural networks which learn the obstacle boundaries and the free space maps, while the symbolic/procedural part is composed of a library of heuristic algorithms which analyze the neural maps on a topological bases, playing the role of “situation recognizers”. An incremental merging algorithm is also presented, which integrates in real time the spatio-temporal information of reconstructed graphs of navigation situations