In such curiosity-driven exploration, the robot is motivated to explore sensorimotor activities for which it is making learning progress: i.e. its estimated performances in prediction or in mastery are increasing (and thus uncertainty is decreasing).
Thus, there is an intrinsic reward for learning progress, managed by a meta‐cognitive structure which then allows the learner to order autonomously its own learning experiences, creating its own curriculum where skills, including the manipulation of external objects, progressively increase in complexity.

In particular, such experiments suggest that the onset of language can spontaneously forms out of such sensorimotor development, where vocalizations are discovered by the learner to be special forms of “tools” that allow to manipulate a special kind of external entities, i.e. others (as a side effect, the meta-cognitive system is itself forming internal distinctive categories of “self” vs. “physical objects” vs. “others”, based on how much they create learnable interactions).

References:
Gottlieb, J., Oudeyer, P‐Y., Lopes, M., Baranes, A. (in press) Information seeking, curiosity and attention: computational and neural mechanisms, Trends in Cognitive Science.

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.
http://www.pyoudeyer.com/ims.pdf

Oudeyer, P‐Y., Kaplan, F. (2006) Discovering communication, Connection Science, 18(2), pp. 189‐‐206.
http://www.pyoudeyer.com/ConnectionSc…

https://flowers.inria.fr/TRO-12-Rouan…

This paper studies the impact of interfaces, allowing nonexpert users to efficiently and intuitively teach a robot to rec- ognize new visual objects. We present challenges that need to be addressed for real-world deployment of robots capable of learning new visual objects in interaction with everyday users. We argue that in addition to robust machine learning and computer vision meth- ods, well-designed interfaces are crucial for learning efficiency. In particular, we argue that interfaces can be key in helping non-expert users to collect good learning examples and, thus, improve the performance of the overall learning system.

Then, we present four alternative human–robot interfaces: Three are based on the use of a mediating artifact (smartphone, wiimote, wiimote and laser), and one is based on natural human gestures (with a Wizard-of-Oz recognition system). These interfaces mainly vary in the kind of feedback provided to the user, allowing him to understand more or less easily what the robot is perceiving and, thus, guide his way of providing training examples differently.

We then evaluate the impact of these interfaces, in terms of learning efficiency, usability, and user’s experience, through a real world and large-scale user study. In this experiment, we asked participants to teach a robot 12 different new visual objects in the context of a robotic game. This game happens in a home-like environment and was de- signed to motivate and engage users in an interaction where using the system was meaningful. We then discuss results that show significant differences among interfaces. In particular, we show that interfaces such as the smartphone interface allows nonexpert users to intuitively provide much better training examples to the robot, which is almost as good as expert users who are trained for this task and are aware of the different visual perception and machine learning issues. We also show that artifact-mediated teaching is significantly more efficient for robot learning, and equally good in terms of usability and user’s experience, than teaching thanks to a gesture-based human-like interaction.

Presentation of the Ergo-Robot Experiment by Pierre-Yves Oudeyer, during “Symposium on Language Acquisition and Language Evolution”, Stockholm University, Royal Academy of Sciences, Stockholm, Sweden (organised by Francesco Lacerda and Bjorn Lindblom).

=== Summary

In a big egg that has just opened, a tribe of young robotic creatures evolves and explores its environment, wreathed by a large zero that symbolizes the “origin.” Beyond their innate capabilities, they are outfitted with mechanisms that allow them to learn new skills and invent their own language. Endowed with artificial curiosity, they explore objects around them, as well as the effect their vocalizations produce on humans. Human, also curious to see what these creatures can do, react with their own gestures, creating a loop of interaction which progressively self-organizes into a new communication system established between man and ergo-robots.

The Ergo-Robot Experiment addresses a very important methodological and experimental need in Developmental Robotics: evaluating how algorithmic architectures for learning and development can scale up to long-term real world robot experiments (i.e. robot operating and learning continuously for several weeks or months). For this goal, which Ergo-Robots addresses [P28], and in the context of a research lab, one needs an experimental platform which is very robust, with precise and reliable motor control, easy to use and maintain, relatively cheap, and reconfigurable (in order to evaluate algorithms on robots with different morphologies, i.e. different robots).

The experiment was presented at the exhibition “Mathematics: A Beautiful Elsewhere” at Fondation Cartier pour l’Art Contemporain (scenography made in collaboration with film director David Lynch) provided in itself a large impact for scientific dissemination of our research activities towards the general public, being explained by student scientific mediators to around 70000 visitors (trained students were continuously present in the exhibition), and being reported multiple times in the general press (e.g. France 2, France Inter, France Culture, RFI, BFM TV, Slate.fr, Sciences et Avenir, New Scientist, Financial Times).

ts software and hardware development was realized by the INRIA ENSTA ParisTech Flowers team in collaboration with University of Bordeaux/Labri: Jérome Béchu, Fabien Bénureau, Haylee Fogg, Paul Fudal, Hugo Gimbert, Matthieu Lapeyre, Olivier Ly, Olivier Mangin, Pierre-Yves Oudeyer, Pierre Rouanet. Scenography was realized in collaboration with David Lynch and his team. Scenography was realiszed in collaboration with David Lynch and his team.

=== Main references

Web site: https://flowers.inria.fr/robots/ergo-robots/

Oudeyer, P-Y. (2011) Curiosity and Languages, in Catalogue of the Exhibition “Mathematics: A Beautiful Elsewhere”, Fondation Cartier pour l’Art Contemporain, Paris, France.
https://flowers.inria.fr/OudeyerCatalogueExpoMathematics2011GB.pdf

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.
http://www.pyoudeyer.com/ims.pdf

Steels, L. (2003) Evolving grounded communication for robots. Trends in Cognitive Science. 7(7), July 2003, pp. 308-312.
http://www.csl.sony.fr/downloads/papers/2003/steels-03c.pdf

Gottlieb, J., Oudeyer, P-Y., Lopes, M., Baranes, A. (2013)
Trends in Cognitive Science, 17(11), pp. 585-596. http://dx.doi.org/10.1016/j.tics.2013.09.001

Keywords: imitation learning, robotics, human behavior understanding, motion primitive, dictionary learning, multimodal learning, dance

We present an approach, based on non-negative matrix factorization, for learning to recognize parallel combinations of initially unknown human motion primitives, associated with ambiguous sets of linguistic labels during training. In the training phase, the learner observes a human producing complex motions which are parallel combinations of initially unknown motion primitives. Each time the human shows a complex motion, he also provides high-level linguistic descriptions, consisting of a set of labels giving the name of the primitives inside the complex motion. From the observation of multi-modal combinations of high-level labels with high-dimensional continuous unsegmented values representing complex motions, the learner must later on be able to recognize, through the production of the adequate set of labels, which are the motion primitives in a novel complex motion produced by a human, even if those combinations were never observed during training. We explain how this problem, as well as natural extensions, can be addressed using non-negative matrix factorization. Then, we show in an experiment in which a learner has to recognize the primitive motions of complex human dance choreographies, that this technique allows the system to infer with good performance the combinatorial structure of parallel combinations of unknown primitives.

More details can be found in the paper :
Mangin O., Oudeyer P.Y., Learning to recognize parallel combinations of human motion primitives with linguistic descriptions using non-negative matrix factorization. IEEE/RSJ International Conference on Intelligent Robots and Systems, Vilamoura, Algarve (Portugal), 2012.
http://olivier.mangin.com/publi.html#…
https://flowers.inria.fr/ManginOudeye…

A formal approach to social learning: Exploring language acquisition through imitation
Presentation made by Thomas Cederborg for his PhD viva, december 2013.

This presentation covers topics presented in details in the following research articles:

From Language to Motor Gavagai: Unified Imitation Learning of Multiple Linguistic and Non-linguistic Sensorimotor Skills
Thomas Cederborg; Pierre-Yves Oudeyer
IEEE Transactions on Autonomous Mental Development (TAMD), IEEE, 2013
https://flowers.inria.fr/CederborgOud…

Simultaneous Acquisition of Task and Feedback Models
Manuel Lopes; Thomas Cederborg; Pierre-Yves Oudeyer
Development and Learning (ICDL), 2011 IEEE International Conference on, 2011, Germany. Development and Learning (ICDL), 2011 IEEE International Conference on, pp. 1 – 7
http://hal.archives-ouvertes.fr/docs/…

Imitating Operations On Internal Cognitive Structures for Language Aquisition
Thomas Cederborg; Pierre-Yves Oudeyer
HUMANOIDS, 2011, Bled, Slovenia. HUMANOIDS Proceedings, pp. NA
http://hal.archives-ouvertes.fr/docs/…

Incremental Local Online Gaussian Mixture Regression for Imitation Learning of Multiple Tasks
Thomas Cederborg; Adrien Baranes; Li Ming; Pierre-Yves Oudeyer
International Conference On Intelligent Robots and Systems (IROS), 2010, Taipei, Taiwan, Province Of China.
https://flowers.inria.fr/Cederborgeta…

Presentation of the Ergo-Robot Experiment by Pierre-Yves Oudeyer, during “Symposium on Language Acquisition and Language Evolution”, Stockholm University, Royal Academy of Sciences, Stockholm, Sweden (organised by Francesco Lacerda and Bjorn Lindblom).

=== Summary

In a big egg that has just opened, a tribe of young robotic creatures evolves and explores its environment, wreathed by a large zero that symbolizes the “origin.” Beyond their innate capabilities, they are outfitted with mechanisms that allow them to learn new skills and invent their own language. Endowed with artificial curiosity, they explore objects around them, as well as the effect their vocalizations produce on humans. Human, also curious to see what these creatures can do, react with their own gestures, creating a loop of interaction which progressively self-organizes into a new communication system established between man and ergo-robots.

The Ergo-Robot Experiment addresses a very important methodological and experimental need in Developmental Robotics: evaluating how algorithmic architectures for learning and development can scale up to long-term real world robot experiments (i.e. robot operating and learning continuously for several weeks or months). For this goal, which Ergo-Robots addresses [P28], and in the context of a research lab, one needs an experimental platform which is very robust, with precise and reliable motor control, easy to use and maintain, relatively cheap, and reconfigurable (in order to evaluate algorithms on robots with different morphologies, i.e. different robots).

The experiment was presented at the exhibition “Mathematics: A Beautiful Elsewhere” at Fondation Cartier pour l’Art Contemporain (scenography made in collaboration with film director David Lynch) provided in itself a large impact for scientific dissemination of our research activities towards the general public, being explained by student scientific mediators to around 70000 visitors (trained students were continuously present in the exhibition), and being reported multiple times in the general press (e.g. France 2, France Inter, France Culture, RFI, BFM TV, Slate.fr, Sciences et Avenir, New Scientist, Financial Times).

ts software and hardware development was realized by the INRIA ENSTA ParisTech Flowers team in collaboration with University of Bordeaux/Labri: Jérome Béchu, Fabien Bénureau, Haylee Fogg, Paul Fudal, Hugo Gimbert, Matthieu Lapeyre, Olivier Ly, Olivier Mangin, Pierre-Yves Oudeyer, Pierre Rouanet. Scenography was realized in collaboration with David Lynch and his team. Scenography was realiszed in collaboration with David Lynch and his team.

=== Main references

Web site: https://flowers.inria.fr/robots/ergo-robots/

Oudeyer, P-Y. (2011) Curiosity and Languages, in Catalogue of the Exhibition “Mathematics: A Beautiful Elsewhere”, Fondation Cartier pour l’Art Contemporain, Paris, France.
https://flowers.inria.fr/OudeyerCatalogueExpoMathematics2011GB.pdf

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.
http://www.pyoudeyer.com/ims.pdf

Steels, L. (2003) Evolving grounded communication for robots. Trends in Cognitive Science. 7(7), July 2003, pp. 308-312.
http://www.csl.sony.fr/downloads/papers/2003/steels-03c.pdf

Gottlieb, J., Oudeyer, P-Y., Lopes, M., Baranes, A. (2013)
Trends in Cognitive Science, 17(11), pp. 585-596. http://dx.doi.org/10.1016/j.tics.2013.09.001

En 1950, dans son article Computing Machinery and Intelligence, Alan Turing proposait pour la première fois l’idée de construire une machine qui apprendrait comme un enfant : « Plutôt que d’écrire un programme qui simulerait le fonctionnement mental d’un adulte, pourquoi ne pas en produire un qui simulerait celui d’un enfant ? Si celui-ci suivait alors une éducation appropriée, alors on pourrait obtenir un cerveau adulte. » Cette idée est restée longtemps presque inexplorée, et ce n’est que depuis une quinzaine d’années qu’elle est devenue le centre d’un nouveau domaine scientifique : la « robotique développementale ». Ce nouveau champ est dédié à l’étude et à la modélisation informatique et mathématique des processus du développement sensorimoteurs et social.

Résumé:
Les systèmes de vocalisations humains, véhicules physiques du langage, sont caractérisés par des formes et des propriétés structurales complexes. Ils sont combinatoriaux, basés sur la ré-utilisation systématique de phonèmes, et l’ensemble des répertoires de phonèmes des langues du monde est marqué à la fois par de fortes régularités statistiques, les universaux, et une grande diversité. En outre, ce sont des codes culturellement partagés par chaque communauté de locuteurs. Quelle est l’origine des formes de la parole ? Quels sont les mécanismes qui, au cours de la phylogenèse et de l’évolution culturelle, ont permis leur évolution ? Comment un code de la parole partagé peut-il se former dans une communauté d’individus ? Je vais m’intéresser dans ce chapitre à la manière dont les phénomènes d’auto-organisation, et leurs interactions avec la sélection naturelle, peuvent permettre d’éclairer ces trois questions.
La tendance qu’ont de nombreux systèmes physiques complexes à générer spontanément des formes nouvelles et organisées, comme les cristaux de glaces ou les spirales galactiques, est présente en effet tout autant dans le monde inorganique que dans le monde vivant. Ainsi, l’explication de l’origine des formes du vivant ne peut reposer uniquement sur le principe de sélection naturelle, qui doit être complémenté par la compréhension des mécanismes de génération de formes nouvelles dans lesquels l’auto-organisation est centrale. Or, ceci s’applique aux formes sociales et culturelles du vivant, en particulier aux formes de la parole et du langage. Je vais donc ainsi commencer par articuler de manière générale les relations entre auto-organisation, sélection naturelle et néo-Darwinisme pour la compréhension de la genèse des formes du vivant. Je vais ensuite instancier ces relations dans le cadre des trois questions que j’ai énoncées ci-dessus. J’expliquerai alors pourquoi l’utilisation de simulations et de modèles informatiques est fondamentale pour faire progresser les théories qui y sont afférentes. Enfin, je présenterai un exemple d’expérimentation d’un modèle informatique qui montre que certains mécanismes simples de couplages sensorimoteurs permettent de générer des systèmes de parole combinatoriaux, caractérisés par la dualité universaux/diversité, et partagés culturellement. Je conclurai par les scénarios évolutionnaires que cette expérimentation informatique vient complémenter ou renouveler.

Oudeyer, P-Y. (2009) L’auto-organisation dans l’évolution de la parole, in Dehaene, S., and Petit, C., Parole et Musique: Aux origines du dialogue humain, Colloque annuel du Collège de France, pp. 83-112, Odile Jacob.
http://www.pyoudeyer.com/ParoleetMusi…

Aux sources de la parole: auto-organisation et évolution
Oudeyer, P-Y. (sept. 2013)
Odile Jacob, Paris.
http://www.pyoudeyer.com/AuxSourcesDe…