Léo Lopez
I2M, Aix-Marseille Université
/user/leo.lopez/
Date(s) : 18/06/2018 iCal
14 h 00 min - 15 h 00 min
One of the long-term goal in robotics is to introduce in human environment autonomous robots capable of helping humans and interacting with them in everyday life safely and efficiently. In a non-controlled environment, a robot, in order to perceive and act must face an incomplete information problem. In this work, we naturally opted for a modeling approach that can quantify that uncertainty: the probabilistic approach. The methods developed here are largely inspired by biological (brain) solutions to problems of prediction, inference and control. All of which have been recently cast within the same Bayesian scheme which is widely used in motor control : predictive coding. In this framework, prediction is the key, estimation of future states of the body or the environment are taken into account for a particular task, not only present and past states. This approach has been applied to explain a wide variety of phenomena: action understanding, perception-action loops and perceptual learning, action selection or goal-directed behavior . We use one of its implementation in this thesis, the Bayesian filtering scheme and we refer in this presentation to two bodies of literature, one in robotics / control and the other one in neuroscience, which use both of them a convergent lexicon (e.g Bayesian inference). We will present several methods for robotics without to resort on the inverse kinematics that could be too time consuming for real-word applications. Our framework (predictive coding and the Bayesian brain hypothesis) and simulations permitted us to adress different questions in cognitive science like the usefulness of the visual and motor space for reaching and grasping and a robotic model of sensorimotor integration.
http://quintonj.free.fr/index.php/Profiles/LopezL
https://www.researchgate.net/profile/Leo_Pio-Lopez
Catégories