Biography

Stephane Doncieux is Professor in Computer Science at the University Pierre and Marie Curie, in the ISIR lab. He got his MSc in 1999 and his PhD in 2003, in the LIP6, the computer science laboratory of UPMC. He has participated to the creation of the Institute of Intelligent Systems and Robotics (ISIR) in 2007 and became, with Bruno Gas, leader of the SIMA research team. The team gathered researchers in mechatronics, signal processing and computer science to build autonomous robots. In 2011, he became leader of the AMAC research team (Architecture and Models of Adaptation and Cognition), an inter-disciplinary  team with 13 permanent members coming from both computer science and computational biology. The team studies learning and adaptation to understand their underlying mechanisms in the living and to design algorithms to provide robots with such abilities. He became professor in 2012. He is currently the coordinator of the DREAM project (FET H2020, http://robotsthatdream.eu/).

Abstract

Bootstrapping development for robots to deal with open environments

How could a robot adapt to an environment that its designer knows little about ? This is a major challenge whose resolution could result in a new generation of robots that would have many applications, from service to manufacturing. Dealing with it implies for the system to build its own knowledge and acquire relevant information for a particular task and domain and learn about how to use it. Humans are faced with this issue and rely on developmental stages in which they progressively acquire the knowledge and skills required to deal with the complexity of their environment. In the frame of the FET H2020 DREAM project, we have defined a formalization of a developmental process together with criteria to measure its progress. The acquisition of knowledge has been decomposed into developmental waves to progressively acquire the knowledge required by the robot to solve the tasks it is facing and transfer the corresponding knowledge to new contexts. A first feature of the project is to rely on evolutionary methods to generate behaviors without the need to provide demonstrations. A second feature is to draw inspiration from the consolidation and restructuring processes occurring during sleep and observed in humans or rats and build a cognitive architecture relying on experimentations on the real robot (“day” processes) and exploration, consolidation or restructuring with the robot turned off (“night” processes). Besides robotics applications, the proposed models could also give insights in the mechanisms at play in the brain. Several neuroscience case studies are thus considered, notably the development of bird song.