Martian World Models: Controllable Video Synthesis with Physically Accurate 3D Reconstructions

arXiv 25.7 2025 62.6 method, system, application

TLDR

A data pipeline and video generator for synthesizing realistic Martian landscapes from real NASA stereo images, achieving high visual fidelity and 3D consistency.

Reasoning

The paper's strengths include its use of real NASA data and a novel pipeline to bridge the domain gap, but its focus is narrow (Martian terrain) and it lacks interactive or reinforcement learning components.

Read-first score

Read-first score 62.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 30.

Methodology quality 25%

100

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

Recency 8%

86.7

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

Reproducibility 25%

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

Topical relevance 42%

42.9

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

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 479.

Keyword Scores

video world model

world model

generative world model

world dynamics prediction

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

M3arsSynth: a data curation pipeline that reconstructs 3D Martian environments from real stereo navigation images (NASA PDS) and renders high-fidelity multiview 3D video sequences at metric-scale resolution.
MarsGen: a Martian terrain video generator fine-tuned on M3arsSynth data, capable of synthesizing videos conditioned on an initial frame, camera trajectories, or textual prompts.
Holistic solution combining physically accurate 3D reconstruction with controllable video synthesis to address Martian data scarcity and domain gap.

Methodology

The approach consists of two components: M3arsSynth, which reconstructs 3D Martian surface models from real stereo navigation images sourced from NASA's Planetary Data System and renders multiview video sequences; and MarsGen, a video generator fine-tuned on this synthetic data to produce novel videos that are visually realistic and geometrically consistent with the 3D structure, conditioned on an initial frame and optionally camera trajectories or text prompts.

Key Results

The proposed method outperforms video synthesis models trained on terrestrial datasets, achieving superior visual fidelity and 3D structural consistency in generated Martian landscape videos.

Limitations

Reliance on stereo navigation images from NASA PDS may limit coverage and resolution for certain Martian terrains.
Fine-tuning on synthetic data from M3arsSynth may introduce a sim-to-real gap when applied to real Martian imagery.
Evaluation is only against terrestrial video synthesis models, not against other Martian-specific approaches.

Resources

Paper Project