⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

Agent Planning with World Knowledge Model

arXiv 24.5 2024 61.9 method

TLDR

Introduces a parametric World Knowledge Model to improve LLM agent planning by providing global prior and local dynamic knowledge, reducing trial-and-error and hallucination.

Reasoning

The paper presents a novel approach to enhance LLM-based agent planning with a World Knowledge Model, showing strong empirical results on real-world simulated datasets. However, the abstract lacks details on the model architecture and limitations, and the term 'world model' is used loosely, potentially overclaiming.

Read-first score

Read-first score 61.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 20.

Methodology quality 25%
90

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,dataset,experiment,metric,result

Reproducibility 25%
85

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=code,dataset,github

Recency 8%
75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Topical relevance 42%
28.6

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

Methodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 730.

Keyword Scores

world model
8
interactive world model
4
world dynamics prediction
3
generative world model
2
model-based reinforcement learning world model
2
world simulator
1
video world model
0

Deep Analysis

Innovations

  • Parametric World Knowledge Model (WKM) providing global prior knowledge and local dynamic knowledge for agent planning
  • Self-synthesis of knowledge from both expert and sampled trajectories
  • Instance-level task knowledge that generalizes better to unseen tasks
  • Weak WKM can guide strong agent model planning
  • Unified WKM training showing promising potential for further development

Methodology

The method steers the agent model to self-synthesize knowledge from expert and sampled trajectories. A parametric World Knowledge Model (WKM) is then developed to provide prior task knowledge for global planning and dynamic state knowledge for local planning. Experiments are conducted on three complex real-world simulated datasets using three state-of-the-art open-source LLMs (Mistral-7B, Gemma-7B, Llama-3-8B) and compared against various strong baselines.

Key Results

The proposed method achieves superior performance compared to strong baselines across all datasets and LLMs. It effectively alleviates blind trial-and-error and hallucinatory action issues, and demonstrates that instance-level task knowledge generalizes to unseen tasks, weak WKM can guide strong agent models, and unified WKM training has promising potential.

Resources

PaperCode

Tags