Engineering Knowledge for Automated Planning

Abstract

Automated planning is a prominent Artificial Intelligence challenge, as well as being a common capability requirement for intelligent autonomous agents. A critical aspect of what is called domainindependent planning, is the application knowledge that must be added to the planner to create a complete planning application. This is made explicit in (i) a domain model, which is a formal representation of the persistent domain knowledge, and (ii) an associated problem instance, containing the details of the particular problem to be solved. Both these components are used by automated planning engines for reasoning, in order to synthesize a solution plan. Formulating knowledge for use in planning engines is currently something of an ad-hoc process, where the skills of knowledge engineers significantly influence the quality of the resulting planning application. On top of that, a notion of quality of the knowledge captured within a domain model is missing; it is therefore hard to provide useful guidelines to knowledge engineers. This paper raises some issues relating to the engineering of application knowledge for automated planning, focussing on the domain model. It uses the idea of a domain model as a formal specification of a domain, and considers what it means to measure the quality of such a specification. To do this it proposes definitions of the attributes of a domain model and its encoding language, which are needed by the automated planning community in order to improve tools for supporting the engineering of planning knowledge, and to advance toward a shared and inclusive definition of quality of domain models.

In Defence of Design Patterns for AI Planning Knowledge Models

“…As noted in the previous section, domain-independent planning systems are heavily dependent on the provided knowledge models; this is also true when explainability is concerned. The behaviour of a planning system can not transcend the knowledge model, and the importance of the planning knowledge model has been well-argued [17]. In fact, in the XAIP field, the knowledge model is more and more regarded as a source of knowledge that can explain the behaviour of the planning system [24,11].…”

Automated Planning Encodings for the Manipulation of Articulated Objects in 3D with Gravity

“…As a side effect of our work, we also designed a set of challenging PDDL+ models, that can be used to compare the performance of planning engines. We also believe they provide a good ground for testing and exploiting knowledge engineering techniques for evaluating the quality of different planning encodings [19]. As future work, we would like to test our approach on different robots models, and to evaluate more PDDL+ planners, namely SMTPlan+ [8], DReal [4] and EZCSP [2].…”

D-VAL: An automatic functional equivalence validation tool for planning domain models

“…Among other tasks, this activity is concerned with checking the correctness of a planning domain model with respect to its specification, requirements or any other reference. If the reference is described informally, then the process of validating the correctness of domain models is also informal [9]. On the other hand, when the requirements are described formally, it is feasible to perform formal validation and to automate this process.…”

“…Two planning domain models are functionality equivalent if both can be used to solve the same set of planning problems. The need for a technique to analyse planning domain models for functional equivalence has been highlighted in the literature [11,12,9]. One example application is the evaluation of the quality of planning model learning algorithms [1,14].…”