Answers are grounded in this paper and cite their sources.
Restart procedures must account for unavailable equipment and the interlocks that constrain action ordering. We specify SHARP, a hierarchical planning architecture with ordering, temporal, resource and interlock annotations, and report a small symbolic comparison. Thirty scenarios over an eight-unit synthetic plant use three action-precondition interlocks and failure-dependent reduced goal sets. The reported campaign compares in-house breadth-first search, A*, greedy best-first search with safety pruning, HTN-style goal regression and SHARP. All methods satisfy the symbolic validator in all scenarios; this is not evidence that the full physical restart mission or continuous safety envelope was restored. SHARP's median decomposition/repair count is 29, versus 472 state expansions for A* and 618 for breadth-first search; these counters measure different objects. Its reported median time is 0.18 ms versus 15.1 ms for A*, while mean plan length is 15.83 versus 14.20 actions. Greedy search expands fewer states (median 20.5), and the reported corrected comparison favors greedy on that counter. Removing interlock-aware ordering raises SHARP's reported median count to 40; the source describes the ablations as derived from logged internals, whose provenance requires verification. Risk-priority removal produces no detected difference. Mid-execution repair, temporal/resource scheduling, continuous trajectory safety and cap-binding scalability remain untested.
No takes yet. Share an insight, caveat, or question.
Kiyadeh et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: