The creation and maintenance of a domain model is a well recognised bottleneck in the use of automated planning; indeed, ensuring a planning engine is fed with an accurate model of an application is essential in order that generated plans are effective. Engineering domain models using a hybrid representation is particularly challenging as it requires accurately describing continuous processes, which can have complex numeric effects. Existing approaches can exploit observation data in order to automatically refine process models, reducing the knowledge engineering effort in producing a detailed process model. However, the usefulness of the resulting model depends on the completeness of the presented data. In particular, datasets are often gathered using default or hand crafted strategies, which can be different from plans generated by planning systems. In this work we consider the data collection problem. We first consider two methods of identifying existing sparse areas in the data. We then consider how plans can be generated that are tailored towards gathering data in these areas. Observations made during the execution of these plans can then be used to supplement the dataset, leading to more accurate and complete domain models as more observations are added. In this paper we present the problem and outline our intended solution.
|Publication status||Published - 20 Dec 2021|
|Event||36th Workshop of the UK Planning and Scheduling Special Interest Group 2021 - Online|
Duration: 20 Dec 2021 → 20 Dec 2021
|Workshop||36th Workshop of the UK Planning and Scheduling Special Interest Group 2021|
|Abbreviated title||UK PlanSIG 2021|
|Period||20/12/21 → 20/12/21|