Position paper: Reasoning about domains with PDDL

One of the major drivers for the progress in scalability of automated planners has been the introduction of the Planning Domain Definition Language (PDDL) and the International Planning Competition (IPC). While PDDL provides a convenient formalism to describe planning problems, there is a significant gap with regards to describing domains. Although PDDL is split into a domain description and a problem description, the domain description is not enough to specify a domain completely, as it does not constrain the possible problems in the domain. For example, there is nothing in the BLOCKSWORLD PDDL domain description which says that a block can not be on top of itself in the initial state. In this position paper, we argue that PDDL domains should be extended to incorporate a new section which constrains possible problems in the domain. We argue that such an extension can be based on first-order logic, and describe several use cases where this extension might be of use. We also provide some preliminary empirical results of one way for automatically extracting such constraints based on mutual exclusion.