My former research focused on the processes of development in infants, in order to both gain insights into human learning and as a method for structuring learning in robotics.
Development in the infant consists of a series of stages, corresponding to learning and consolidation of abilities, such as reaching, crawling and walking. The stages are defined by constraints, which may be cognitive, sensorimotor, anatomical, or maturational, but which all act to limit the abilities and actions of the infant at various times. These constraints do not negatively affect the development of the infant, but rather prevent over-extension. They effectively act to shape learning, limiting interactions and reducing the perceived complexity of the environment.
Even for robots with limited sensing and actuating capabilities the complexity arising from interactions with the real world makes unconstrained learning unreliable and computationally demanding. By progressing through a sequence of staged development, similar to that seen in infancy, a robot will be able to overcome these problems. Motor babbling and play provide intrinsically-generated mechanisms to drive learning.
These videos below show work undertaken during the IM-CLeVeR project. The first explains how the iCub robot learns visually-triggered reaching using a developmental process. The second shows the iCub learning about objects and interactions using play-like behaviour.
Law, J., Shaw, P., Earland, K., Sheldon, M., & Lee, M. H. (2014). A psychology grounded approach for longitudinal development in cognitive robotics. Frontiers in Neurorobotics, 8:1. doi: 10.3389/fnbot.2014.00001.