Incremental Planning for Procedural Models in Aerie

Planning for spacecraft can be computationally intensive, especially when including higher-fidelity geometry and attitude control. If only a small change is made to the inputs, the resulting generated plan may not change much. Thus, it is desirable to compute a new plan from a prior one and avoid recomputing parts that do not change. We refer to this as incremental planning. There are several automated planning systems that can do incremental planning naturally based on declarative and restrictive modeling languages where dependencies among activities and state variables are explicit. We show how this can be done for models specified procedurally in Java for the Aerie planning system. The challenge is that activity behavior is a black box, and dependencies are only partially seen while generating the plan. We describe details of the Aerie plan generation algorithm and how it is modified for incremental planning. We validated the approach on a planning model for the Europa Clipper spacecraft used in operations and discuss additional challenges in performance and usability.