Towards learning strategies and exploration patterns for feature perception

Abstract

During infancy, infants spend a lot of time visually exploring the scene around them. Over the first year of life, the level of detail that can be perceived visually increases significantly. In this study, the ability to perceive areas of interest w.r.t. human developmental change in vision, specifically acuity and field of view over the first year of life, is investigated. Two scenarios, namely learning through a series of developmental changes and learning without any constraints, shed light on how a humanoid robot scaffolds learning of interesting areas in the scene through different emergent exploratory behaviours. Divergence/convergence in features is reported, demonstrating a potential to be used at a higher level of understanding. Staged strategies with early sensory constraints and exploratory behaviour based on “similarity searches” improve the quality of acquired features and may be used as a mechanism for better on-line learning of objects knowledge.