Per Aage Brandt                            Oct. 17, 2008

The architectural design of a robot is a primordial and crucial problem. A simple and classic design could look like this: perception -> calculus -> motor function. Perception: scanning and creating ‘input’. Motor function creating ‘action’. A robot is an embodied computer; so a cognitive robot is an embodied computer with cognitive functions. All cognitive functions build on representation. The design of such a device could thus be:  perception -> calculus 1 -> representation -> cognition -> calculus 2 -> motor function. There has to be a pre-representational calculus (1) and a post-cognitive calculus (2), because only a programmed calculus can run representation and motor function.

    By representation, we will – because we must – understand a ‘third-person’ (meaning: viewpoint neutral) space topographically invested with limits, stationary and mobile objects, ego-position (mobile), and voids (open regions allowing motion), that is, a map. In this spatial representation – the map – the position of ego (the subject robot) is of course determined relative to the topography, not by the subjective vantage point of the ego (the ‘self’ of the robot). However, the ego can review, that is, can scan, the representation, that is, read the map; this function corresponds to basically ‘remembering’ the external ‘world’.

    By intentional cognition we will mean a set of systems generating behavior that ‘makes sense’ as motivated and based on ‘projects’ for action; such project systems can be narrative or epistemic: aiming at changing a situation or aiming at knowing a situation. In both cases, we will have to install a project space containing a project schema. This schema diagrammatically shows the change intended, or the filling of a knowledge gap intended. The project space can show the project purely and ideally, as an ideal narrative program, written into a void representation. The advantage of the ideal program is of course to be independent of particular circumstances. But then it has to be translated into a realistic narrative program by projection into the remembered map of the actual circumstances. The result of this projection is an intentional representation (IR) that can feed the motor calculus. So IR should fill the cognition slot in the flow shown above.

    A linguistic narrative will now be within reach. The intentional plan, according to the realistic narrative program, is interpreted by the calculus as a process divided by metric ‘steps’. It is sequenced, so that it can stop after each step and perform a feedback scan in real space. This allows a re-adjustment of the map and thus of the realistic plan, namely if the outer world has changed in some way. For each such step, a ‘narrative’ situation occurs. The term episode will refer to the situation where a re-adjustment does occur due to an external change. If nothing has changed, the situation is non-dramatic and thus non-episodic. The series of episodes generated by the completion of an action program is a story. The linguistic account of each episode in the story and of the way in which the episodes refer backwards to each other (by anaphor) will generate the text of the story.

    All dialogical devices involving user and robot will build upon and be grounded in this narrative capacity of the cognitive robot.

*