Planning Engine — References

Planning formalisms, prompting schemes that produce explicit strategies, and the evidence about how well language models actually plan.

Prompting schemes that materialise a strategy

• Plan-and-Solve Prompting (Wang et al., 2023), arXiv:2305.04091 — separating strategy formation from step execution within a single prompt, and the measurable effect of doing so.
• Tree of Thoughts: Deliberate Problem Solving with Large Language Models (Yao et al., 2023), arXiv:2305.10601 — search over candidate intermediate states with an explicit evaluator, the reference design for strategy search.
• HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face (Shen et al., 2023), arXiv:2303.17580 — planning as the production of a typed, dependency-bearing task list bound to a concrete catalogue of executors.
• ReWOO: Decoupling Reasoning from Observations — plans the full step sequence before any tool output is seen, to avoid re-sending accumulated observations. Cited by name.

Classical planning

• STRIPS: A New Approach to the Application of Theorem Proving to Problem Solving (Fikes and Nilsson, 1971) — preconditions, effects, and goal states; the vocabulary every later planner borrows. Cited by name.
• PDDL — the Planning Domain Definition Language — the standard encoding for planning domains and problems, and the target format when a model is used to translate a goal for a classical solver. Cited by name; no single canonical URL is recorded.

Evidence about model planning ability

• On the Planning Abilities of Large Language Models (Valmeekam et al.) — evaluation on classical planning benchmarks, and the case for external verification of generated strategies rather than trust. Cited by name; the preprint identifier is not recorded here.