Section: Partnerships and Cooperations

International Initiatives

INRIA International Partners

• Luis Montesano, University of Zaragoza, Spain. Manuel Lopes collaborated with Luis Montesano on active learning approaches for grasping point learning. Results were published in Robotics and Autonomous Systems [16] .

• Francisco Melo Instituto Superior Técnico, Portugal. Manuel Lopes collaborated with Francisco Melo on the development of active learning for inverse reinforcement learning. Recent developments consider the extension to more cues available to the learner and sampling complexity on the algorithm.

• José Santos-Victor, Instituto Superior Técnico, Portugal. Manuel Lopes collaborated with José Santos-Victor on the extension of affordances models to higher levels of representations, e.g. relational models.

• Maya Cakmak, Andrea Thomaz, Georgia Tech, USA. Manuel Lopes collaborated with Maya Cakmak on the development of optimal teaching algorithms for sequential decision problems (modeled as markov decision processes). The algorithm provides optimal demonstrations for systems that learn using inverse reinforcement learning. The joint work considers not only the algorithmic aspects but also a comparison with human behavior and the possibility of using insights from the algorithm to elicit better teaching behavior on humans [41] .

• Marc Toussaint, Tobias Lang, Free University of Berlin, Germany. Manuel Lopes and Pierre-Yves Oudeyer are collaborating with FUB in the unification of exploration algorithms based on intrinsic motivation with methods for exploration in reinforcement learning such as $R_{max}$. We intend to develop a general framework for exploration in non-stationary domains. Another project consider how to learn efficient representation for robotic hierarchical planning.

• Jacqueline Gottlieb, Columbia University, New-York, US. Adrien Baranes, Pierre-Yves Oudeyer and Manuel Lopes began a collaboration with Jacqueline Gottlied, neuroscientist at Columbia University and specialist of visual attention and exploration in monkeys. An experimental set-up with brain imaging and behavioural observations of monkeys, and made to evaluate new families of computational models of visual attention and exploration (some of which developped in the team around the concept of intrinsic motivation) is being elaborated. Adrien Baranes will go in postdoc at Jacqueline Gottlieb's laboratory through a FullBright grant, and experiments shall begin next year.

• Louis ten Bosch, Radboud University, The Netherlands. Pierre-Yves Oudeyer and David Filliat continued to work with Louis ten Bosch on the modelling of multimodal language acquisition using techniques based on Non-Negative Matrix Factorization. We showed that these techniques can allow a robot to discover audio-video invariants starting from a continuous unlabelled and unsegmented flow of low-level auditory and visual stimuli.A journal article is in preparation.

• Paul Vogt (Tillburg University, The Netherlands), Linda Smith (Indiana University, Bloomington, US), Aslo Ozyurek (Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands), Tony Belpaeme (University of Plymouth, UK). Pierre-Yves Oudeyer began collaboration with partners of the NWO SCMSC project to set up a research network on modeling of social cognition and symbolic communication.

• Michael Gienger from Honda Research Institute Europe. Alexander Gepperth collaborated with Principal Scientist Dr.Michael Gienger from Honda Research Institute Europe GmbH about robotic grasping: this activity will result in a jointly supervised internship ("stage de fine d'études") and a publication.

• Ursula Korner from Honda Research Institute Europe. Alexander Gepperth collaborated with Senior Scientist Ursula Korner from Honda Research Institute Europe GmbH on the topic of "Biologically motivated models of robotic memory acquisition and consolidation using the PROPRE algorithm". This activity has resulted in the submission of a conference publication to the European Symposium on Neural Networks (ESANN) 2012 and will result in additional journal publications as well as the creation of a robotic demonstration system.

• Michael Garcia Ortiz, Laboratory for Cognitive Robotics (CoR-Lab) in Bielefeld, Germany. Alexander Gepperth collaborated with Michael Garcia Ortiz, a PhD student from the Laboratory for Cognitive Robotics (CoR-Lab) in Bielefeld, Germany, on the exploitation of scene context for object detection in intelligent vehicles. This collaboration resulted in the submission of a journal publication to the journal "Neurocomputing".

• Martha White, Patrick Pilarski, Joseph Modayil, Adam White, and Richard Sutton, University of Alberta, Canada. Thomas Degris is collaborating with Martha White, Patrick Pilarski, Joseph Modayil, Adam White, and Richard Sutton from the Reinforcement Learning and Artificial Intelligence group at the University of Alberta on new learning algorithms for robots. One paper is in the process of being published, two others are work in progress. Moreover, via the University of Alberta, Thomas Degris uses for his research a cluster belonging to the Alberta Ingenuity Centre for Machine Learning, and a cluster from Westgrid (http://www.westgrid.ca/ ), a member of the High Performance Computing consortia in Canada.

• Stefan Schaal, University of Southern California (Los Angeles, USA). Freek Stulp is continuing his collaborative work with Prof. Stefan Schaal of the University of Southern California (Los Angeles, USA), and founding director of the Max-Planck-Insitute for Intelligent Systems (Tübingen, Germany). This project aims at combining algorithms from evolutionary optimization and direct reinforcement learning to achieve adaptive exploration for life-long learning in a developmental robotics context. We intend to submit this work the the International Conference on Intelligent Robots and Systems as well as the International Conference on Development and Learning.

Visits of International Scientists

• Marc Toussaint, Technical University Berlin, Germany (november 2011)

• Tobias Lang, Technical University Berlin, Germany (november 2011)

• Luis Montesano, University of Zaragoza, Spain (november 2011)

• Yukie Nagai, Osaka University, Japan (march 2011)

• Jan Peters, TU Darmstadt, Germany (september 2011)

• Robert Damper, University of Southampton, UK (may 2011)

• Stefano Nolfi, University La Sapienza, Roma, Italy (may 2011)

• Jacqueline Gottlieb, Columbia University, US (december 2011)

• Thomas Degris, University of Alberta, Canada (may 2011)

• Thierry Chaminade, Mediterranean Institute for Cognitive Neuroscience (INCM), CNRS - Aix-Marseille Universite) (may 2011)

• Ludovic Marin, University of Montpellier, France, (may 2011)

• Robin Salesse, University of Montpellier, France, (may 2011)

• Olivier Sigaud, University Paris VI (may 2011)

Participation In International Programs

NWO project: Socio-Cognitive Mechanisms of Symbolic Communication

SCMSC is a project funded by NWO (Netherlands Organization for Scientific Research) on Socio-Cognitive Mechanisms of Symbolic Communication, and coordinated by Paul Vogt (Tillburg University). This project aims to study the socio-cognitive mechanisms of symbolic communication. In contrast to other species, humans have the capacity to communicate symbolically (i.e. using forms that are either arbitrary or conventionalised) in an open fashion (i.e. with a very large repertoire of symbols). It is widely accepted that our ability to communicate symbolically has both cognitive and social roots. In recent years, traditional approaches from humanities to study symbolic communication, such as linguistics and psychology, have been complemented by computational approaches. However, interactions between researchers from the humanities with computer modellers have been few and far between, perhaps due to a lack of mutual understanding of what each field can contribute to the other.

In this project, we will set up a structural research network to improve cross fertilization between researchers from different disciplines by exchanging knowledge and experiences, and join forces to study communication multidisciplinary. This way, we aim to improve each other’s research methods and investigate unifying properties of the socio-cognitive mechanisms underlying symbolic communication. To achieve this, we propose to start up an open structural research network in which we will organise two workshops, apply for joint research funding, set up an online repository of publications, educational and other materials, and publish an edited collection.

Partners are Paul Vogt (Tillburg University, The Netherlands), Linda Smith (Indiana University, Bloomington, US), Pierre-Yves Oudeyer (INRIA-ENSTA-ParisTech, France), Aslo Ozyurek (Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands), Tony Belpaeme (University of Plymouth, UK). Web site: http://ilk.uvt.nl/~paul/scmsc/SCMSC/Home.html .