I am research director (DR1) at Inria, heading the Flowers team at Inria Bordeaux Sud-Ouest (see PhD students). I was previously a permanent researcher in Sony Computer Science Laboratory for 8 years (1999-2007).
I study lifelong autonomous learning, and the self-organization of behavioural, cognitive and language structures, at the frontiers of artificial intelligence, machine learning, and cognitive sciences. I use machines as tools to understand better how children learn and develop, and I study how one can build machines that learn autonomously like children, within the new field of developmental artificial intelligence.
With colleagues, I develop fundamental computational theories of curiosity and intrinsically motivated learning (with AI, neuroscience and developmental psychology perspectives). We proposed the learning progress hypothesis to explain key aspects of human spontaneous exploration in high-dimensional bodies and environments. We showed that such forms curiosity-driven exploration can self-organize long term developmental trajectories, accouting for how infants progressively develop vocal skills, tool use and language.
From an AI and machine learning perspective, we study how machine can efficiently acquire world models, as well as open-ended repertoires of skills over an extended time span. In particular, we develop deep reinforcement learning agents able to learn to represent and generate their own goals, leveraging grounded language learning for out-of-distribution creative exploration (see this blog post), and automatic curriculum learning. This addresses the major AI challenge of how to learn autonomously in high-dimensional environments, when there are no external rewards and many potential distractors. We are combining these approaches with self-supervised deep learning techniques, used to learn spaces in which to self-generate goals, to discover independently controllable features, solve efficiently sparse reward problems in Deep RL, and learn efficiently modular goal-conditioned policies. We also study how neuro-symbolic architectures can enable learning structured representations and handling systematic compositionality and generalization. Recently, we started exploring the new area of automated discovery of self-organized patterns in complex systems, leveraging intrinsically motivated goal exploration and unsupervised representation learning.
I also work on theoretical models of the origins and evolution of speech and language, studying the role of self-organization in neural networks and agents dynamical coupling. In the new edition of my book “Self-organization in the evolution of speech” (to appear in 2020 at OUP, CC-BY), I present an integrated view of the roles of self-organization and intrinsic motivation in the origins of language.
Applications: societal and industrial impact. I also work on real-world applications grounded in this fundamental research. In educational technologies, we use curiosity-driven learning algorithms to personalize sequences of exercises for human learners, maximising learning efficiency and motivation: after an initial series of experiments with >1000 children in primary schools, we are now working with a consortium of edTech companies and the support of French ministry of Education to integrate this approach in an educational software for wide dissemination. We developed open-source educational robotics kits (some used by dozens of thousands of children in the world and adapted in largely disseminated educational books e.g. from Main à la Pâte), now disseminated by a non-governmental organization and a start-up. In robotics, our curiosity algorithms have been used within the Sony Aibo and Qrio humanoid entertainment robots. I also worked on emotional speech synthesis technologies used in some Playstation video games, and various forms of adaptive human-computer interfaces.
Link to information about the Poppy humanoid robot : Poppy Project web site. Poppy is an open-source 3D printed robot for science, education and art designed by the Flowers team. It was built to study the impact of the body on sensorimotor development and cognition: it makes it possible to really consider the body as an experimental variable. See article at Humanoids 2013 conference.
I collaborate regularly with artist within project that explore the frontiers between art and science. This has been the opportunity to create original connections between the general public and our scientific projects, in particular by bringing people to ask themselves and to ourselves stimulating questions about the position of such scientific projects within society at large.
Examples of such projects include:
Our now blog post is out! See Language as a Cognitive Tool: Dall-E, Humans and Vygotskian RL Agents
I have received in october 2018 the Inria Prize of the National Science Academy (computer science researcher under 40).
The team is actively contributing to the large scale Adaptiv’Maths edTech project, coordinated by evidenceB and supported by the French ministry of education, transferring ZPDES machine learning technique we developed (see B. Clement’s PhD) to personalize sequences of maths exercises to be used large scale in French schools and beyond.
Akakzia, A., Colas, C., Oudeyer, P. Y., Chetouani, M., & Sigaud, O. (2021, May). Grounding Language to Autonomously-Acquired Skills via Goal Generation. In ICLR 2021.
Colas, C., Karch, T., Lair, N., Dussoux, J. M., Moulin-Frier, C., Dominey, P., & Oudeyer, P. Y. (2020). Language as a Cognitive Tool to Imagine Goals in Curiosity Driven Exploration. Advances in Neural Information Processing Systems (Neurips 2020), 33.
Etcheverry, M., Moulin-Frier, C., & Oudeyer, P. Y. (2020, December). Hierarchically Organized Latent Modules for Exploratory Search in Morphogenetic Systems. In NeurIPS 2020-34th Conference on Neural Information Processing Systems.
Portelas, R., Colas, C., Weng, L., Hofmann, K., & Oudeyer, P. Y. (2020). Automatic Curriculum Learning For Deep RL: A Short Survey. In Proceedings of IJCAI.
Invited talks at the ICRA 2021 workshops on “Learning to learn for robotics” and “Towards curious robots: modern approaches for intrinsically motivated intelligent behavior”
Invited talk at ELLIS workshop on Meta-Learning, on Instrinsically motivated learning and automatic curriculum learning as meta-learning.
Invited seminar at the MIT embodied AI seminar, on developmental machine learning, Deep RL and artificial curiosity, April 2020, https://
Keynote talk at the Crossmodal Learning Center Autumn School, on Developmental Machine Learning, Curiosity and Deep RL, Dec. 2020. https://
Invited talk at the Brain and Cognition seminar at the University of Geneva, on Curiosity-driven learning in humans and mahcines, Oct. 2020. https://
Keynote talk at the EGC conference in Brussels, on developmental machine learning, Jan. 2020, https://
See the Jobs page on the Flowers website.
Les robots et l’intelligence artificielle (Nathan, 2020)
Mondes Mosaïques (CNRS Editions, 2016)
Aux sources de la parole (Odile Jacob, 2013)
Self-organization in the evolution of speech (Oxford University Press, 2006, updated version 2020)
The Economist, sept. 2018: A sense of curiosity is helpful for artificial intelligence; Scientific American, 2018: Intelligent Machines that Learn Like Children; Pour la Science, 2015: L’éveil des bébés robots La recherche, 2015 Le roboticien des sciences humaines Le Monde, nov. 2014 (portrait): Pierre-Yves Oudeyer, aussi curieux que ses robots Socialter, sept. 2014: Les robots seront-ils aussi “bêtes” que nous? Des machines et des hommes Les Echos, mars 2014 : Les robots auront un impact sur la société