When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems
The paper discusses the phenomenon of epistemic miscalibration in planning within LLM-based multi-agent systems. It highlights how agents can misjudge their knowledge, leading to failures even when actions are executed correctly. The authors propose a new workflow, EPC-AW, to improve planning accuracy and system-level success rates.
- ▪Epistemic miscalibration occurs when agents misjudge their knowledge during planning, leading to potential failures.
- ▪The proposed Epistemic Planning Calibration Agentic Workflow (EPC-AW) assesses plan support under varying information conditions.
- ▪Experiments indicate that EPC-AW enhances system-level success by an average of 9.75%.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.23414 (cs) [Submitted on 22 May 2026] Title:When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems Authors:Zehao Wang, Shilong Jin, Zhao Cao, Lanjun Wang View a PDF of the paper titled When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems, by Zehao Wang and 3 other authors View PDF HTML (experimental) Abstract:LLM-based multi-agent systems can fail even when planned actions are executed correctly because agents may misjudge their knowledge when evaluating plan feasibility, a phenomenon we term epistemic miscalibration in planning.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
