ASCoL: Automated Acquisition of Domain Specific Static Constraints from Plan Traces
Rabia Jilani, Andrew Crampton, Diane E. Kitchin, Mauro Vallati · University of Huddersfield Repository (University of Huddersfield) · 2014
Domain-independent planning systems require that domain constraints and invariants are specified as part of the input domain model. In AI Planning, the generated plan is correct provided the constraints of the world in which the agent is operating are satisfied. Specifying operator descriptions by hand for planning domain models that also require domain specific constraints is time consuming, error prone and still a challenge for the AI planning community. The LOCM (Cresswell, McCluskey, and West 2013) system carries out automated generation of the dynamic aspects of a planning domain model from a set of example training plans. We enhance the output domain model of the LOCM system to capture static domain constraints from the same set of input training plans as used by LOCM to learn dynamic aspects of the world. In this paper we propose a new framework ASCoL (Automated Static Constraint Learner), to make constraint acquisition more efficient, by observing a set of training plan traces. Most systems that learn constraints automatically do so by analysing the operators of the planning world. Out proposed system will discover static constraints by analysing plan traces for correlations in the data. To do this an algorithm is in the process of development for graph discovery from the collection of ground action instances used in the input plan traces. The proposed algorithm will analyse the complete set of plan traces, based on a predefined set of constraints, and deduces facts from it. We then augment components of the LOCM generated domain with enriched constraints.