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

Reason--Imagine--Act: Closed-Loop LLM Decision Making with World Models for Autonomous Driving

arXiv 2026 60.3 method, application

TLDR

A closed-loop LLM framework with a world model for online safety verification in autonomous driving, outperforming baselines in CARLA.

Reasoning

The paper presents a novel integration of LLM reasoning with world model rollouts for real-time safety, supported by strong empirical results in simulation. However, it lacks real-world validation and the world model is not generative or video-based, limiting the scope of some keyword scores.

Read-first score

Read-first score 60.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 36.

Recency 6%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%
81.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.81579843

Reproducibility 18%
81

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

Methodology quality 18%
60

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

Topical relevance 29%
51.4

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

Citation velocity 12%
0

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 463.

Keyword Scores

world model
9
world dynamics prediction
8
interactive world model
6
world simulator
5
model-based reinforcement learning world model
5
generative world model
2
video world model
1

Deep Analysis

Innovations

  • Closed-loop framework coupling LLM reasoner with action-conditioned world model for online safety verification
  • Action template and candidate sub-actions proposed by LLM, with short-horizon rollouts by world model and safety scorer selection
  • Feedback from safety scorer to next reasoning step, enabling dynamic verification of physical feasibility

Methodology

The Reason-Imagine-Act (RIA) framework integrates an LLM reasoner with an action-conditioned world model. At each decision step, the LLM generates an action template and candidate sub-actions; the world model performs short-horizon rollouts to predict outcomes; a safety scorer selects the safest executable action and provides feedback to the LLM for subsequent reasoning. The framework is evaluated under a unified CARLA point-goal protocol with 1000 episodes, comparing against training-free baselines CARLA TM and MADA on closed-loop metrics.

Key Results

RIA achieves 80.05% route completion, 51.10% arrival rate, and 0.20% collision rate, consistently outperforming training-free baselines under the same closed-loop interface.

Tags