Qwen-RobotWorld Technical Report: Unifying Embodied World Modeling through Language-Conditioned Video Generation

arXiv 2026 67.3 method, system

TLDR

Qwen-RobotWorld is a language-conditioned video world model for embodied intelligence, unifying multiple domains via video generation and achieving top benchmark results.

Reasoning

Strengths include a unified formulation across diverse embodied tasks, a large-scale dataset (EWK), and strong benchmark performance. Weaknesses are limited discussion of limitations or failure cases, and reliance on benchmarks may not fully capture real-world complexity.

Read-first score

Read-first score 67.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 57.

Recency 6%

100

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

Citation impact 18%

94.5

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.94455183

Topical relevance 29%

81.4

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

Methodology quality 18%

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 406.

Keyword Scores

world model

video world model

generative world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

interactive world model

Deep Analysis

Innovations

Unified language-conditioned video world model for embodied intelligence across robotic manipulation, autonomous driving, indoor navigation, and human-to-robot transfer using natural language as a unified action interface.
Double-Stream MMDiT with MLLM Action Encoding: a 60-layer double-stream diffusion transformer coupling frozen Qwen2.5-VL semantics with video-VAE latents through layer-wise joint attention.
Embodied World Knowledge (EWK) dataset: 8.6M video-text corpus (200M+ frames) with action-language mapping over 20+ embodiments and 500+ action categories.
General+Expert Progressive Curriculum: two-stage training strategy that first learns general visual priors then injects embodied specialization under a shared language interface.
Three application directions: synthetic data generation for policy training augmentation, scalable virtual environments for policy evaluation, and language-guided planning signals for downstream robot control.

Methodology

The model employs a Double-Stream MMDiT architecture with 60 layers, coupling frozen Qwen2.5-VL semantics with video-VAE latents via layer-wise joint attention. It is trained on the Embodied World Knowledge (EWK) dataset (8.6M video-text pairs, 200M+ frames, 20+ embodiments, 500+ action categories) using a two-stage General+Expert Progressive Curriculum: first learning general visual priors, then injecting embodied specialization under a shared language interface.

Key Results

The model ranks 1st overall on EWMBench and DreamGen Bench, and outperforms all open-source models on WorldModelBench and PBench. Zero-shot analyses on RoboTwin-IF benchmark demonstrate robust generalization and multi-view consistency.

Resources

Paper