Theories of psychophysics draw clear parallels between choice behaviour and statistical hypothesis testing, the latter being a computational approach that can be easily implemented on a mobile robot. A learner can benefit from modifying a decision criterion (threshold) for detecting changes, or paying attention, to adapt its own measure of significance to particular subjective situations. Social cues from a teacher agent simplify the job of attention by exposing the learner to parts of the environment relevant to the task. Using results from simulated and physical experiments we show the role of threshold adaptation, and of the utilisation of social cues, in robotic imitation.
Due to copyright, the preprint is not available on-line. You can request a copy by e-mailing me.