On augmenting reactivity with deliberation in a controlled manner

Abstract. We argue that a reactive agent obeying the requirement of predictability imposed by a hard-real-time application domain cannot be equipped with an arbitrarily powerful deliberation capability as this would jeopardize the predictability of the agent's behaviour. Therefore such augmentation should be performed in a principled, controlled man-ner. We illustrate our line of thought with the example of Generic Lay-ered Architecture used for creating reactive agents acting in dynamic environments requiring real-time responsiveness.