Recognition and classification of path features with self-organizing maps during reactive navigation

The paper focuses on recognition and classification of path features during navigation of a mobile robot. The extracted features play the role of relevant navigation situations as (in a corridor), (at a turning point), (in a narrow passage). The method is an incremental learning and classification technique, based on a self-organizing neural model. Two different self-organizing networks are used to encode occupancy bitmaps generated from sonar patterns in terms of obstacles boundaries and free paths, and heuristic procedures are applied to these growing networks to add and prune units, to determine topological correctness between units, to distinguish and categorize features.