Applying self-organizing networks to recognizing rooms with behavior sequences of a mobile robot

We describe the application of a self-organizing network to the robot which learns to recognize rooms (enclosures) using behavior sequences. In robotics research, most studies on recognizing environments have tried to build the precise geometric map with highly sensitive sensors. However many natural agents like animals recognize the environments with low sensitivity sensors, and a geometric map may not be necessary. Thus we attempt to build a mobile robot using a self-organizing network to recognize the enclosures, in which it acts, with low sensitivity and local sensors. The mobile robot is behavior-based and does wall-following in an enclosure. Then the sequences of behaviors executed in each enclosure are obtained. The sequences are transformed into real-value vectors, and inputted to the Kohonen self-organizing network. Unsupervised learning is done and a mobile robot becomes able to distinguish and identify enclosures. We fully implemented the system using a real mobile robot and made experiments for evaluating the ability. Consequently we found out the recognition of enclosures was done well and our method was robust against small obstacles in an enclosure.