Novelty Detection on a Mobile Robot Using Habituation

  Request Code

  API Access   Call/Text an Expert

In this paper a novelty filter is introduced which allows a robot operating in an un structured environment to produce a self-organised model of its surroundings and to detect deviations from the learned model. The environment is perceived using the rob ot's 16 sonar sensors. The algorithm produces a novelty measure for each sensor scan relative to the model it has learned. This means that it highlights stimuli which h ave not been previously experienced. The novelty filter proposed uses a model of hab ituation. Habituation is a decrement in behavioural response when a stimulus is pre sented repeatedly. Robot experiments are presented which demonstrate the reliable o peration of the filter in a number of environments.