Learn, plan, remember: A developmental robot architecture for task solving

Abstract

This paper presents a robot architecture heavily inspired by neuropsychology, developmental psychology and research into 'executive functions' (EF) which are responsible for the planning capabilities in humans. This architecture is presented in light of this inspiration, mapping the modules to the different functions in the brain. We emphasize the importance and effects of these modules in the robot, and their similarity to the effects in humans with lesions on the frontal lobe. Developmental studies related to these functions are also considered, focusing on how they relate to the robot's different modules and how the developmental stages in a child relate to improvements in the different modules in this system. An experiment with the iCub robot is compared with experiments with humans, strengthening this similarity. Furthermore we propose an extension to this system by integrating with 'Epigenetics Robotic Architecture' (ERA), a system designed to mimic how children learn the names and properties of objects. In the previous implementation of this architecture, the robot had to be taught the names of all the necessary objects before plan execution, a learning step that was entirely driven by the human interacting with the robot. With this extension, we aim to make the learning process fully robot-driven, where an iCub robot will interact with the objects while trying to recognise them, and ask a human for input if and when it does not know the objects' names.