Living beings are often observed switching strategies in response to a changing environment. However, autonomous robotics mostly implements a single behaviour well suited to a particular task such as navigation, localization and so on. Actually, one burning issue of autonomous robotics is to manage a complex task starting from a set of simple behaviours. In other words, the robot has to choose the optimal behaviour given the sensori-motor context in order to build a global and coherent process. This is usually done by a strict specification from the programmer. In this article, we put forward a framework called behaviours hierarchy that handle elementary ability to respond to a given task. We show that this framework leads to the continuous application of an adequate behaviour depending on the environment. Finally, we propose a general method to implement this framework using Bayesian programming.