Towards High-Consistency Embodied World Model with Multi-View Trajectory Videos

arXiv 25.11 2025 63.2 method

TLDR

MTV-World uses multi-view trajectory videos for high-consistency embodied world model visuomotor prediction.

Reasoning

The paper introduces a novel multi-view trajectory video control method to address spatial information loss in embodied world models, and proposes an auto-evaluation pipeline. Strengths include addressing a key limitation of low-level action translation and using multi-view for consistency. Weaknesses are the lack of explicit real-world validation details and potential complexity of multi-view setups.

Read-first score

Read-first score 63.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 52.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%

74.3

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 25%

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

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 255.

Keyword Scores

world model

video world model

generative world model

world dynamics prediction

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Using trajectory videos obtained through camera intrinsic/extrinsic parameters and Cartesian-space transformation as control signals instead of low-level actions
Multi-view framework to compensate for spatial information loss from projecting 3D actions onto 2D images
Auto-evaluation pipeline leveraging multimodal large models and referring video object segmentation models, with Jaccard Index for spatial consistency

Methodology

MTV-World forecasts future frames based on multi-view trajectory videos as input, conditioning on an initial frame per view. It employs camera intrinsic/extrinsic parameters and Cartesian-space transformation to convert low-level actions into trajectory videos, and uses a multi-view framework to mitigate spatial information loss from 2D projection.

Key Results

Extensive experiments demonstrate that MTV-World achieves precise control execution and accurate physical interaction modeling in complex dual-arm scenarios.

Resources

Paper