Explicit knowledge and the deliberative layer: Lessons learned

Abstract

Over the last four years, we have been slowly ramping up explicit knowledge representation and manipulation in the deliberative and executive layers of our robots. Ranging from situation assessment to symbolic task planning, from verbal interaction to event-driven execution control, we have built up a knowledge-oriented architecture which is now used on a daily basis on our robots.This article presents our design choices, the articulations between the diverse deliberative components of the robot, and the strengths and weaknesses of this approach. We show that explicit knowledge management is not only a convenient tool from the software engineering point of view, but also pushes for a different, more semantic way to address the decision-making issue in autonomous robots. I. A KNOWLEDGE-ORIENTED ARCHITECTUREA. Towards the cognitive robot at LAAS Natural interaction and cooperation are on the (dare we say, short-term) agenda for the human-robot interaction community. They are keys to the broad class of interactive manipulation problems: several agents agree on a (more or less implicit) joint goal that requires some sort of cooperation to be successfully achieved. This class of problems involves both dialogue and manipulation and they are often initially underspecified: they require iterative and interactive resolution.Over the last years we have focused our efforts on identifying the cognitive prerequisites of these challenges, and giving them experimental reality on the robots: what is required for sentences like "Let's set the table together" to be understood by the robot, correctly interpreted in the spatial and temporal context of the interaction, and eventually transformed into a set of actions.We have chosen to tackle the challenge from several ends: human-aware navigation and motion planning [1], situation assessment coupled with motion planning [2], projection of "mightabilities" that anticipates what surrounding agents may do [3], and the design and deployment of a pervasive knowledge-oriented software architecture.The last item is the focus of this paper: what "knowledgeoriented" architecture means, and what its benefits and drawbacks are. B. Scope of the articleIn essence, this article portrays our use of explicit semantic interfaces to integrate several deliberative components into a decisional architecture for our robots. It reports on our experience with acquiring, managing and reusing grounded knowledge in the context of human-robot interaction.Unlike previous publications by the authors that introduced the deliberative modules of our robots independently from each other, this contribution is an account of the importance of explicit knowledge manipulation when integrating them into a large cognitive architecture.While we do not present experiments in this paper, the ideas and techniques we present all have been implemented and tested on several robots (see previously reported experiments in [4]-[7], [9]). Videos presenting some of these experimental results can be watched online at http://www.laas.fr/...

Intentional joint agency: shared intention lite

“…Our approach has been informed by our experience with the BDI model of agency [5] and several associated agent architectures -architectures that were introduced to support a balance of deliberative and reactive behaviours, and that in their instantiation are reliant on domain-specific expert knowledge acquisition to provide a knowledge level view [6], c.f. [7], [8]. We are also supporters of the position that logicbased techniques are well suited to represent social reasoning and through which to engineer effective mechanisms, c.f.…”

“…Even though our focus is on cognitive mechanisms as essential components of an integrated cognitive architecture for effective social robots, and we have some exploratory work on human communication patterns [39], we recognise there are many topics important in such architectures that we do not attempt to address -spatial reasoning [40], dialogue actions [7], multimodal inputs [41], action signalling [24], the link between perception and action [42], [43], and comparisons between logic-based reasoning and other approaches such as game theory [44] and probabilistic reasoning [45] ... to name but a few! !…”

Graph-based software knowledge: Storage and semantic querying of domain models for run-time adaptation

“…In fact, Niemueller et al [23] have shown that it is feasible to apply a document-oriented database like MongoDB, even for logging raw sensor data and analyzing robots' behavior in retrospect. Also knowledge-enabled and ontologybased approaches such as KnowRob [5], Robo Brain [6] or the OpenRobot Ontology (ORO) [7] rely on knowledge bases to store and query specifications of robots, their capabilities, tasks and environments. For the matter, this is complementary to the graph databases which we propose in this paper.…”

“…This involves a) persistently storing the different notations and formats of DSLs; b) composing the various domain models; and c) querying over multiple domains at run time. The AI community has already realized this requirement, as evidenced by knowledge-enabled approaches like KnowRob [5], RoboBrain [6] or the OpenRobot Ontology (ORO) [7]. At the core of these approaches, graph-based knowledge representations such as ontologies provide common representations and query interfaces to the robot's runtime environment.…”

FourByThree: Imagine humans and robots working hand in hand

“…The integration of plan synthesis and continuous plan execution has been demonstrated both for timeline based planning (e.g., [21]) and PDDL based (e.g., [22]). In scenarios of human robot interaction important problems have been addressed: (a) ''human aware'' planning has been explored for example in [23], (b) the interaction of background knowledge for robotic planning in rich domain (addressed for example in [24], (c) synthesis of safety critical plans to guarantee against harmful states (relevant in co-presence with humans) is addressed in [25] and [26]). Within the FourByThree project, a timeline-based planning approach is pursued relying on the APSI-TRF [27], developed for the European Space Agency and exploited in several missions.…”