I. INTRODUCTION
The paper examines the question of how to create robots able to learn incrementally and online a variety of new context-dependant tasks whose number and complexity is not known at programming time, and when the demonstrator is also not allowed to tell the system how many tasks there are and whether a given demonstration corresponds to a variation of an already demonstrated task or to a new tasks. Moreover, the demonstrator may alternate demonstrations corresponding to different tasks in an uncontrolled or even random order. Yet, we assume that elements of the sensorimotor context can allow the robot to statistically infer information that may allow it to accurately reproduce the right task in a given sensorimotor context.