Producing Parameterized Value Functions Through Modulation For Cognitive Developmental Robots

Parameterizing value functions as a representation of robotic tasks in different domains allows for their generalization, and can provide a way to transfer knowledge to new situations. To this end, in this paper we propose a modulation based mechanism embedded within a cognitive architecture for robots. It makes use of the combined operation of the long-term memory and the motivational system in order to select candidate primitive value functions for transfer. These are then adapted to the new situation through the addition of modulatory ANNs to progressively conform new parameterized value functions able to address more complex situations in a developmental manner. The proposed method is tested in a Baxter robot, which must solve different tasks in a cooking setup.