Perceptual Homing by an Autonomous Mobile Robot using Sparse Self-Organizing Sensory-Motor Maps

We present a biologically-motivated approach to the problem of perception-based homing by an autonomous mobile robot. A three-layered self-organizing network is used to autonomously learn the desired mapping from perceptions to actions. The network, which bears some similarities to the structure of the mammalian cerebellum, is initially trained by teleoperating the robot on a small number of homing paths within a given domain of interest. During training, the connections between input sensory layer and the hidden layer as well as those between the hidden layer and the motor output layer are modiied according to the well-known competitive Hebbian learning rule. By employing a population averaging scheme for computing output motor vectors, the robot can subsequently home from arbitrary locations within the domain based solely on current perceptions. We describe preliminary results based on simulation for an actual mobile robot, equipped with simple photoreceptors and infrared receivers, navigating within an enclosed obstacle-ridden arena.