Biography

Gianluca Baldassarre received the B.A. and M.A. degrees in economics and the M.Sc. degree in cognitive psychology and neural networks from the University of Rome “La Sapienza,” Rome, Italy, in 1998 and 1999, respectively, and the Ph.D. degree in computer science with the University of Essex, Colchester, U.K., in 2003, with a focus on planning with neural networks. He was a Post-Doctoral Fellow with the Italian Institute of Cognitive Sciences and Technologies, National Research Council, Rome, researching on swarm robotics, where he has been a Researcher, since 2006, and coordinates the Research Group that he founded called the Laboratory of Computational Embodied Neuroscience. From 2006 to 2009, he was a Team Leader of the EU Project “ICEA—Integrating Cognition Emotion and Autonomy” and the Coordinator of the European Integrated Project “IM-CLeVeR— Intrinsically-Motivated Cumulative-Learning Versatile Robots,” from 2009 to 2013, and is currently Team Leader of the EU Project “GOAL-Robots – Goal-based Open-ended Autonomous Learning Robots”. He has over 100 international peer-review publications. His cur- rent research interests include cumulative learning of multiple sensorimotor skills driven by extrinsic and intrinsic motivations. He studies these topics with two interdisciplinary approaches: with computational models constrained by data on brain and behavior, aiming to understand the latter ones and with machine-learning/robotic approaches, aiming to produce technologically useful robots.

Abstract

Self-generated goals and intrinsic motivations as means to acquire repertoire of skills in open-ended learning autonomous robots

When intrinsic motivations started to be employed to foster skill open-ended learning in robots, they were mostly used to directly support the autonomous acquisition of skills. In particular they were used to directly produce a reward signal to train the reinforcement learning of skills. Now it is however becoming clear that this approach gives limited possibilities to support open-ended learning of skills. An approach to face this problem, followed by the European Project GOAL-Robots (Goal-based Open-ended Autonomous Learning Robots) is to foster the autonomous acquisition of skills through the self-generation of goals. Goals are intended as internal representations of desired possible future states of the world that can be re-activated internally and can support the learning and performance of the skills that lead to their accomplishment. In the presentation I will discuss the features and utility of goals for open-ended learning of skills by comparing robot architectures using goals with “reactive” architectures not using goals, the problem of goal self-generation based on intrinsic motivations and other mechanisms, and some possible features of the overall robot architectures needed to employ goals to learn skill repertoires.