Intrinsic Motivations (i.e motivations not connected to rewardrelated stimuli) drive humans and other biological agents to autonomously learn different skills in absence of any biological pressure or any assigned task. In this paper we investigate which is the best learning signal for driving the training of different tasks in a modular architecture controlling a simulated kinematic robotic arm that has to reach for different objects. We compare the performance of the system varying the IntrinsicMotivation signal and we show how a Task Predictor whose learning process is strictly connected to the competence of the system in the tasks is able to generate the most suitable signal for the autonomous learning of multiple skills.
Intrinsic motivation signals for driving the acquisition of multiple tasks: A simulated robotic study
Contributo in atti di convegno
12th International Conference on Cognitive Modelling (ICCM 2013), pp. 59–64, Ottawa, Canada, 11-14 July 2013