Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Domain Engineering to Represent Human Behavior Using Multi-Agent Planning and Inductive Methodologies
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 a mixed domain-engineering methodology that combines multi-agent social protocols, question-asking annotation, procedural grounded theory coding, and HTNs in HDDL to represent human planning behavior in urban search and rescue.
Problem
Automated planning domains often abstract away the nuance, redundancy, and inefficiency of human planning behavior, leading to inaccurate models of collaborative human problem solving in domains such as search and rescue.
Approach
Derive HTN domain structures inductively from annotated question-asking events in SAR videos using procedural grounded theory and Baldoni's social protocol framework, instead of assuming a strategy or goal, and encode them in HDDL.
Key results
- Three annotators labeled 12 SAR videos with an unweighted Cohen's Kappa of 0.892.
- Teams asking questions at a consistent rate scored higher than teams asking the same number sporadically.
- Annotation took 13:30 man-hours combined for three annotators and twelve observations.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| ask questions at a consistent rate rather than sporadically | SAR teams asking the same number of questions · Urban search and rescue | score | Helped |
| ask more questions compared with using other forms of natural-language communication | SAR teams · Urban search and rescue | points | Helped |
| ask questions to clarify an idea rather than instruct each other | SAR teams · Urban search and rescue | collaboration | Helped |
| ask questions to clarify an idea rather than instruct each other | SAR teams · Urban search and rescue | victims rescued | Helped |
| ask questions to clarify an idea rather than instruct each other | SAR teams · Urban search and rescue | points | Helped |
| coordinate on tasks that do not require collaboration | SAR teams · Urban search and rescue | points | Helped |
| coordinate on tasks that do not require collaboration | SAR teams · Urban search and rescue | question-asking behavior | Helped |
Surprising findings
- All teams demonstrated redundancy, nuance, and inefficiency in their planning behavior.
Limitations
- Evaluation limited to annotator agreement; the planned model comparisons and empirical planning evaluation are future work.
- Small sample of 12 videos from a single domain.
Experimental setup
Datasets: urban search and rescue (SAR) video recordings of human teams (12 videos)
Metrics: unweighted Cohen's Kappa
Builds on: Baldoni multi-agent social protocols Procedural Grounded Theory HDDL
Closest papers by meaning
- Large Language Models as Planning Domain Generators (Student Abstract) (AAAI 2024 · similarity 0.85)
- T1: A Tool-Oriented Conversational Dataset for Multi-Turn Agentic Planning (NeurIPS 2025 · similarity 0.85)
- Leveraging Environment Interaction for Automated PDDL Translation and Planning with Large Language Models (NeurIPS 2024 · similarity 0.84)
- PARTNR: A Benchmark for Planning and Reasoning in Embodied Multi-agent Tasks (ICLR 2025 · similarity 0.84)
- PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization (ICLR 2024 · similarity 0.84)
- AutoManual: Constructing Instruction Manuals by LLM Agents via Interactive Environmental Learning (NeurIPS 2024 · similarity 0.83)
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.