Louis Hugues and Alexis Drogoul
This paper addresses the problem of learning robot behaviors in real environments. Robot behaviors have not only to be grounded in the physical world but also in the human space where they are suppose to take place. The paper briefly presents a learning model relying on teaching by demonstrations, enabling the user to transmit its intentions during real experiments. Important properties are outlined and the probabilistic learning process is described. Finally, we indicate how grounded behavior could be interfaced to a symbolic level.