Our paper "Efficient Online Interest-Driven Exploration for Developmental Robots" by Rania Rayyes, Heiko Donat and Jochen Steil, has been accepted at IEEE Transactions on Cognitive and Developmental Systems. The paper proposes new online data-drive learning techniques to reduce the high sample complexity required by the intrinsic motivation methods and developmental robot learning. The efficiency of our proposed learning scheme has been demonstrated on 7 DoF Baxter.
The paper also proposes a novel learning scheme to learn several robot models online, from scratch, simultaneously in "Learning while Behaving" fashion and driven by the robot curiosity to select what to learn and where to discover. The independent-samples t-tests has shown clearly the significant difference in performance accuracy, robustness, and efficiency of our methods compared to the state-of-the art methods