Developmental Robotics

PRESS NEWS

DEVELOPMENTAL ROBOTICS AND

CURIOSITY- DRIVEN LEARNING

NEWS !

I organized a dialog column in IEEE Autonomous Mental Development Newsletter, 4(1), around the question "How can one assess open-ended development?". It is available at: http://www.cse.msu.edu/amdtc/amdnl/AMDNL_V4_N1.pdf

Can a robot be curious? How can a robot learn new tricks by itself and continuously during its lifetime? What mechanisms explain open-ended cognitive development? How can children development inspire robot builders, and how robotic models can be useful to developmental psychology? These are the questions that we study in the developmental robotics group at Sony CSL Paris, which was set up by Frédéric Kaplan and myself.

Studying development, in children or in cognitive robots, is intrinsically difficult because of the complex interplay between embodiment, learning mechanisms and environmental dynamics. A relevant integrative approach can be pursued by viewing development as a complex system the dynamics of which can be studied with embodied models. In particular, together with my colleague Frédéric Kaplan, we conduct research on intrinsic motivation systems (also called intrinsic value systems), that can drive a robot to continuously try to master new know-how. The aim is to construct machines implementing such general capacities as artificial curiosity.

This approach might not only help us understand the mechanisms underlying human development, but it might also provide radically new techniques for building intelligent robots. Indeed, as opposed to the work in classical artificial intelligence in which engineers impose pre-defined anthropocentric tasks to robots, the techniques we develop endow the robots with the capacity of deciding by themselves which are the activities that are maximally fitted to their current capabilities. Our developmental robots autonomously and actively choose their learning situations, thus beginning by simple ones and progressively increasing their complexity. No tasks are pre-specified to the robots, which are only provided with an internal abstract reward function. For example, in the case of the Intelligent Adaptive Curiosity which we developped, this internal reward function pushes the robot to search for situations where its learning progress is maximal.

In particular, we have built an experimental setup, called the Playground Experiment, which allowed to show how the curiosity algorithm which we developped allows for the self-organization of developmental trajectories with sequences of behavioural stages of increasing complexity.

Keywords: developmental robotics, epigenetic robotics, intrinsic motivation, curiosity, values, development, intrinsically motivated reinforcement learning, autonomy, behaviour, developmental trajectory, complexity, active learning.

Selected papers on this subject:

Oudeyer P-Y, Kaplan , F. and Hafner, V. (2007) Intrinsic Motivation Systems for Autonomous Mental Development, IEEE Transactions on Evolutionary Computation, 11(2), pp. 265--286. DOI: 10.1109/TEVC.2006.890271 bibtex reference

Kaplan, F. and Oudeyer, P-Y. (2007) Un robot motivé pour apprendre : le rôle des motivations intrinsèques dans le développement sensorimoteur, Enfance, 1, pp. 46--58.

Oudeyer P-Y., Kaplan F. (2006) Discovering Communication, Connection Science, 18(2), pp. 189--206. bibtex reference

Oudeyer P-Y., Kaplan F., Hafner, V., Whyte A. (2005) The Playground Experiment: Task-Independent Development of a Curious Robot, to appear in the proceedings of the AAAI Spring Symposium Workshop on Developmental Robotics.

Kaplan, F. and Oudeyer, P-Y. (2007) The progress-drive hypothesis: an interpretation of early imitation. In Dautenhahn, K. and Nehaniv, C., editor, Models and mechanisms of imitation and social learning: Behavioural, social and communication dimensions, pp.361--377, Cambridge University Press. 2005.