Planning with Semantic Attachments

Hierarchical Task Networks (HTN) generate plans using a decomposition process guided by extra domain knowledge to guide search towards a planning task. While many HTN planners can make calls to external processes (e.g. to a simulator interface) during the decomposition process, this is a computationally expensive process, so planner implementations often use such calls in an ad-hoc way using very specialized domain knowledge to limit the number of calls. Conversely, the few classical planners that are capable of using external calls (often called semantic attachments) during planning do so in much more limited ways by generating a fixed number of ground operators at problem grounding time. In this paper we develop the notion of semantic attachments for HTN planning using semi co-routines, allowing such procedurally defined predicates to link the planning process to custom unifications outside of the planner. The resulting planner can then use such co-routines as part of its backtracking mechanism to search through parallel dimensions of the statespace (e.g. through numeric variables). We show empirically that our planner outperforms the state-of-the-art numeric planners in a number of domains using minimal extra domain knowledge.