Hierarchical Task Networks (HTN) are a "divide-and-conquer" tool for planning. They organize large, complex problems into several more easily solvable subproblems. The resultant plans also present unique opportunities with regards to plan repair and execution recovery due to the additional structure they possess relative to classical planning.

Many real-world problems are impractical to solve without temporal planning. Temporally interesting problems are in general effectively unsolvable by atemporal planners. We suspect that even temporally simple problems may benefit from a temporal algorithm. Many important real-world problems are dynamic, large, and temporally interesting. Given these qualities we would like to use temporal HTN planning with repair. However, to our knowledge, no existing plan repair algorithm can effectively repair temporal HTN plans.

To date, we have developed a novel formalism for HTN plan repair that is applicable to multiple existing algorithms. Additionally, we have designed and implemented the new HTN repair algorithm IPyHOPPER. Our existing work shows that three state-of-the-art HTN plan repair algorithms have substantially different definitions of what constitutes an acceptable repair. We have also proven how these different notions overlap and differ. Our experiments have evaluated how these algorithms vary in performance over multiple planning domains and problems.

Our proposed research builds on our existing work by considering time. Towards this end we will need to develop a formalism for repairing temporal HTNs, as none exists to our knowledge. We will design and implement the first temporal HTN repair algorithm. This work will extend our theories and proofs regarding atemporal HTNs. Additionally, we will empirically evaluate our temporal HTN algorithm implementation. For this evaluation we will also characterize a temporal plan stability metric. We anticipate that our research will improve the performance of planning systems in dynamic online settings.