The neural autopilot model posits that habit mode relies on retrieving the previous action and its absolute reward prediction error rather than re-evaluating options