GEM-4D: Geometry-Enhanced Video World Models for Robot Manipulation

arXiv 2026 60.4 method, application

TLDR

GEM-4D injects 4D correspondence supervision into video world models for geometrically consistent video prediction and robot manipulation, improving real-world success from 61% to 81%.

Reasoning

The paper addresses a key limitation of video world models—lack of physical grounding—by incorporating dense 4D correspondence supervision, achieving state-of-the-art results in both simulation and real-world manipulation. Strengths include a novel geometry-grounded approach with no extra inference cost and clear real-world improvement; weaknesses are not explicitly discussed in the abstract but may include reliance on a pretrained geometry foundation model.

Read-first score

Read-first score 60.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

Recency 6%

100

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

Citation impact 18%

82.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.82585356

Topical relevance 29%

77.1

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=metric,result

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 456.

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

Injecting dense 4D correspondence supervision distilled from a pretrained geometry foundation model into the video generative backbone during training, enabling joint appearance and geometric structure capture with a single-stream architecture and no additional inference cost.
Introducing an inverse dynamics module that converts correspondence-consistent video rollouts into executable robot trajectories for direct deployment in real-world and simulated manipulation.

Methodology

GEM-4D uses a pretrained geometry foundation model to distill dense 4D correspondence supervision into a video generative backbone during training, allowing the model to jointly capture appearance and geometric structure while maintaining a single-stream architecture with no extra inference cost. An inverse dynamics module is then employed to convert the generated correspondence-consistent video rollouts into executable robot trajectories, enabling direct deployment in both real-world and simulated manipulation tasks.

Key Results

GEM-4D achieves state-of-the-art performance on both video prediction and geometric consistency across simulation and realistic scenarios, and improves real-world manipulation success from 61% to 81%.

Resources

Paper