Embedded Files
Journée d'automne du GT8 Robotique et Neurosciences
Mardi 26 novembre 2019
Equipe Mnemosyne, Inria, Bordeaux
Le groupe de travail "Robotique & Neurosciences" (GT8) du GDR Robotique organise une journée d'automne à Bordeaux. L'objectif est de faire se rencontrer roboticiens, modélisateurs (neurosciences computationnelles, IA, machine learning) et expérimentateurs (neurobiologie et sciences cognitives en général) afin de favoriser les échanges et les collaborations.
Cette journée sera organisée autour d'un ensemble de présentations orales par des chercheurs invités, ainsi que de présentations orales courtes et de posters des participants.
Il y a encore des places libres pour des présentations orales !
Inscription gratuite mais obligatoire (inscriptions fermées).
IN ENGLISH
The "Robotics and Neuroscience" working group (GT8) of the GDR Robotics is organizing a working day this fall to allow roboticians, neuroscientists, psychologists and people working in machine learning (AI) to meet around common themes.
This day will be organized around a series of oral presentations by invited researchers, as well as short oral presentations and posters of the participants
Free registration but mandatory (registration closed).
Organisateurs (Organizers) : Xavier Hinaut, Mehdi Khamassi, Ghilès Mostafaoui, Céline Teulière.
Intervenants invités (Invited speakers) :
Nicolas Navarro Guerrero (Dep. of Eng., Aarhus University, Denmark) : "The Role of Computational Cognitive Science in Autonomous Assistive Systems"
Olivier Ly (Rhoban team, Labri, Bordeaux, FR) : RoboCup - Ligue Humanoid Kid-Size : Des robots Humanoïdes Autonomes qui jouent au foot
Clément Moulin-Frier (Flowers team, Inria Bordeaux, FR) : "Rôle des processus sensori-moteurs, cognitifs et environnementaux dans l’émergence de la communication entre agents"
Bhargav Teja Nallapu (Mnemosyne team, Inria Bordeaux, FR) : "A closed loop framework of decision-making and learning in primate prefrontal circuits : Using virtual experimentation"
Programme (Program) :
9:00-9:25 Accueil et récupération des badges
9:30-9:45 Présentation de la journée (Presentation of the day)
9:45-10:30 Clément Moulin-Frier (Flowers, Inria, Bordeaux): "Self-organization of communication systems in learning agents"
10:30-10:50 Eva Ansermin (Euromov, Montpelier): "Modélisation d'un Contrôle Rythmique Sensorimoteur pour une tâche de passage d'objet coordonnée entre un Homme et un robot."
10:50-11:35 Coffee break + Posters
11:35-11:55 Aymar de Rugy (INCIA, Bordeaux): "Stratégie de contrôle bio inspirée d’une prothèse robotique de bras avec vision artificielle"
11:55-12:40 Nicolas Navarro-Guerrero (Department of Engineering, Aarhus University, Denmark): "The Role of Computational Cognitive Science in Autonomous Assistive Systems"
12:40-14:10 Lunch + Posters
14:10-14:55 Bhargav Teja-Nallapu (Mnemosyne, Inria, Bordeaux): A closed loop framework of decision-making and learning in primate prefrontal circuits : Using virtual experimentation
14:55-15:15 Ghilès Mostafaoui (Ensea, Univ. Cergy-Pontoise): "Coordination interpersonnelle non-intentionnelle et Interface Homme-Machine intuitive"
15:15-16:00 Coffee break + Posters
16:00-16:20 Céline Teulière (UCA, Institut Pascal, Clermont-Ferrand): "Within Reach? Learning to touch objects without prior models"
16:20-17:05 Olivier Ly (Rhoban, Labri, Bordeaux): "RoboCup - Ligue Humanoid Kid-Size : Des robots Humanoïdes Autonomes qui jouent au foot"
17:05-17:30 Concluding remarks and discussions
Résumés (Abstracts of invited talks) :
Title: The Role of Cognitive Robotics in Autonomous Assistive Systems
Abstract: The dream and need for autonomous robots beyond the assembly line are coming together. This development is the product of many factors including a demographic change in all developed countries, deemed the silver tsunami, improvements on computing and machine learning techniques, the pervasive dumb, dangerous, and dirty tasks, etc. Despite the advances in machine learning and automation, there are numerous challenges we need to overcome before developing robots that can operate autonomously alongside people. These involve making robots smarter and capable, but also robots would have to be able to interact with humans naturally and effortlessly. Such capabilities are not only indispensable for domestic robot applications but are also essential to guarantee safe and efficient operation in the production line. In this talk, I will discuss my current and ongoing research on Cognitive Robotics and Human-Robot Interaction towards creating smarter and safer robots.
Title: Self-organization of communication systems in learning agents
Abstract: What are the conditions for complex communication systems to emerge in populations of artificial agents? I propose to address this question by dividing the problem into three subparts. First, there is a sensory-motor grounding problem: How to learn internal representations of the world through the sensory-motor interaction of an agent with its surrounding environment. Second, there is a social grounding problem: How a shared communication system can emerge from local interactions between non-communicating agents. Third, there is a functional grounding problem: How communication can emerge as a way to solve concrete cooperative tasks.
I will address these three aspects through the presentation of three of my previous contributions: (1) Emergence of communicative sound systems in deictic agent populations ; (2) Self-organization of vocal development through curiosity-driven learning ; (3) Language acquisition in human-robot interaction. Finally, since I am just been recruited as a new permanent research in the Flowers group at Inria, I will present my research project which attempts at integrating the three aforementioned aspects of communication in a unified framework. The project relies on recent advances in multi-agent reinforcement learning and conceives communication as the emergent result of a joint behavior optimization process for the realisation of complex cooperative tasks.
Title: A closed loop framework of decision-making and learning in primate prefrontal circuits : Using virtual experimentation
Abstract: How do we make choices between options that are inherently incomparable ? How do we learn to adapt our criteria of decisions from one context to another ? The answer lies in understanding the fundamental value-based decision making systems in the brain, that render our behavior flexible and adaptive, through several interacting neural mechanisms. There are numerous brain regions that contribute in different cognitive processes which emerge into a flexible animal behavior. I will discuss some crucial aspects of the anatomy and functional organization of these brain regions that provide insights into building a cognitive framework to study the flexible behavior. Studying such an adaptive behavior using computational models also requires more dynamic and complex experimentation environments. I will present how the videogame Minecraft has been exploited to serve as a virtual testbed for neuro-computational models of decision-making and learning, thanks to the AI experimentation Malmo platform provided by Microsoft. Finally, I will discuss a specific case of decision-making and the neural pathways implied in the process, to demonstrate how detailed neuro-computational models can be plugged into the high-level cognitive framework of flexible animal behavior.
Title: RoboCup - Ligue Humanoid Kid-Size : Des robots Humanoïdes Autonomes qui jouent au foot
Abstract: Au travers de cet exposé, je donnerai une vue globale de l'initiative RoboCup. Je ferai un focus sur la ligue des robots humanoïdes. J'exposerai l'architecture de ces robots et les grandes problématiques des épreuves. Ensuite, je détaillerai les technologies que nous mettons en oeuvre.
Posters :
Silvia Pagliarini (Inria, Bordeaux) : "A Vocal Learning Model: Development And Analysis"
Which components are needed to build a minimal sensorimotor learning model? Which type of architecture can be used to model imitative learning? Which choices make the model biologically plausible? We propose a general schema for a sensorimotor learning model: sensory space, motor space and perceptual space are linked via a learning framework, a motor control model and a sensory system. We focus on the case of song learning in birds to propose a possible implementation for the motor control and a possible definition of the perceptual space.
Lise Aubin (Euromov/Etis, Cergy): "Socially assistive robots for physical activity engagement in schizophrenia"
Schizophrenia patients (SZ) are more sedentary than healthy individuals with only 25% engaging in the European recommended amount of physical activity (PA). Increasing PA is fundamental to prevent many of physical health problems prevalent in SZ [1]. The goal of this study is to increase PA by manipulating interpersonal coordination between SZ and robots controlled by a neural model rhythmically entrained by human movements. Bidirectional entrainment can then be obtained leading agents to perform synchronous joint actions. Using a robot (Nao) has the advantage of modifying and controlling it without being influenced by the patients as a human would. Patients are instructed to jointly perform fitness movements with the robot. Nao was either synchronized bidirectionally or unidirectionally as a leader with the patient. We will present our first results on non-intentional synchronization between SZ and robot. Patients exhibit a greater motivation to perform tasks with the robot, which open new perspectives in health care protocols. [1] Mitchell AJ, Vancampfort D, Sweers K, van Winkel R, Yu W, De Hert M. Prevalence of Metabolic Syndrome and Metabolic Abnormalities in Schizophrenia and Related Disorders—A Systematic Review and Meta-Analysis. Schizophrenia Bulletin. 2013;39(2):306-318.
Anthony Strock (Inria, Bordeaux) "Using conceptors to transfer between long-term and short-term Memory"
We introduce a model of working memory combining short-termand long-term components. For the long-term component, we used Conceptors in order to store constant temporal patterns. For the short-term component, we used the Gated-Reservoir model: a reservoir trained to hold a triggered information from an input stream and maintain it in a readout unit. Wecombined both components in order to obtain a model in which information can go from long-term memory to short-term memory and vice-versa.
Lieu (Place) :
Enseirb Mat Meca, accès par l'avenue des Universités (200m de l'arrêt de Tram "Arts et Métiers"), 33400 Talence
Amphi D, ancien bâtiment, face Nord (Auditorium D, old building, North part)
Présentez-vous à l'accueil de l'Enseirb MatMéca (bâtiment avec des colonnes et un grand logo "Bordeaux INP").
Un badge avec votre nom vous sera remis. Vous pourrez ensuite suivre les indications pour aller à l'amphi D (côté Nord, ancien bâtiment) où aura lieu la conférence.
Voici le lien pour trouver l'accueil du bâtiment : https://g.page/BordeauxINP?share
Coordonnées GPS :
44°48'20.4"N 0°36'17.7"W
44.805672, -0.604904
#Partage (sharing) :
Lien court de la page (page short link): http://bit.ly/GT8-bdx2019
Sur les réseaux sociaux, utilisez le hashtag #GT8day
Soutiens (partners) :
CNRS, GDR Robotique, Bordeaux INP, Université de Bordeaux.
Plus d'informations sur le groupe de travail GT8 "Robotique et Neurosciences" ICI.
Page updated
Google Sites
Report abuse