Learning from Demonstration by Averaging Trajectories (LAT) has been implemented using a Katana robotic manipulator. LAT works in three phases: Demonstration: Repeated recording of trajectories and detection of involved objects (located at changing positions). Generalisation: Creation of a continuos behaviour model on trajectory level for each object involved. Reproduction: Fusion of the models, adapted to an arbitrary new situation where all the objects again can be located at different positions.