Research map / Large language models / Agents, code and math
Does Robin Hood Use a Lightsaber?: Automated Planning for Storytelling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes combining automated planning (PDDL domain and problem files, planner output as action sequence) with LLM text generation so that each plan step scaffolds a more coherent story, with preliminary child-story experiments and a human study.
Problem
LLM-generated stories lack logical coherence and are prone to repetition and stilted language, and LLMs cannot reason logically about the worlds they describe.
Approach
Use a classical planner over a PDDL domain/problem to produce a valid action sequence, then feed the plan steps one at a time to an LLM so the generated narrative follows a logically valid plan; the paper also outlines future extensions to prompt design, temporal planning and non-deterministic choose-your-own-adventure stories.
Key results
- Almost all nouns (characters, objects, locations) and verbs of the plan are reflected in the generated stories, per PoS analysis.
- Human evaluators rated plan-plus-LLM stories higher than LLM-only stories on metrics such as avoiding repetition and plausible event order.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add a valid automated plan to LLM story generation | LLM storytelling · Children’s stories | human evaluation | Helped |
| add a valid automated plan to LLM story generation | LLM storytelling · Children’s stories | repetition avoidance | Helped |
| add a valid automated plan to LLM story generation | LLM storytelling · Children’s stories | plausibility of event order | Helped |
| use an automated plan to guide LLM story generation | LLM storytelling · Children’s stories | plan noun and verb reflection | Helped |
Limitations
- Preliminary study on a small set of simple children's stories.
- Evaluation is informal: human questionnaire of a small group and PoS overlap, with no full statistics reported in this paper.
- Temporal and non-deterministic extensions are future work only.
Experimental setup
Datasets: The Way Home for Wolf Robin Hood and the Golden Arrow The Paper Bag Princess
Metrics: goal achievement (character and author goals) part-of-speech noun and verb overlap between plan and story questionnaire ratings of story quality
Baselines: LLM-only story generation
Builds on: PDDL TattleTale
Closest papers by meaning
- Planning in the Dark: LLM-Symbolic Planning Pipeline Without Experts (AAAI 2025 · similarity 0.85)
- LLMs Can Plan Only If We Tell Them (ICLR 2025 · similarity 0.84)
- Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researchers (ICLR 2025 · similarity 0.83)
- Large Language Models as Planning Domain Generators (Student Abstract) (AAAI 2024 · similarity 0.83)
- Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks (ICML 2025 · similarity 0.82)
- Agents' Room: Narrative Generation through Multi-step Collaboration (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.