TeleWorld: Towards Dynamic Multimodal Synthesis with a 4D World Model

arXiv 25.12 2025 53.4 method, system

TLDR

TeleWorld proposes a real-time 4D world model unifying video generation, dynamic scene reconstruction, and long-term memory via a generation-reconstruction-guidance paradigm.

Reasoning

The paper introduces a novel framework with hierarchical planning and distillation for real-time synthesis, addressing long-horizon consistency and interaction. However, the abstract lacks explicit real-world experiments or benchmarks, and the approach is described as a report without empirical validation.

Read-first score

Read-first score 53.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 8%

86.7

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

Topical relevance 42%

62.9

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

Methodology quality 25%

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

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

Frontier

Rank sensitivity

Stability: volatile; rank range: 461.

Keyword Scores

world model

generative world model

video world model

interactive world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Generation-reconstruction-guidance paradigm for closed-loop 4D world modeling
Macro-from-Micro Planning (MMPL) for hierarchical planning to reduce error accumulation
Distribution Matching Distillation (DMD) for real-time synthesis under practical computational budgets
Unified 4D framework integrating dynamic object modeling and static scene representation

Methodology

TeleWorld is a real-time multimodal 4D world modeling framework that unifies video generation, dynamic scene reconstruction, and long-term world memory. It employs an autoregressive diffusion-based video model enhanced with Macro-from-Micro Planning (MMPL) and Distribution Matching Distillation (DMD) to achieve real-time synthesis. The system operates in a generation-reconstruction-guidance paradigm where generated video streams are continuously reconstructed into a dynamic 4D spatio-temporal representation that guides subsequent generation to maintain spatial, temporal, and physical consistency.

Key Results

Extensive experiments demonstrate that TeleWorld achieves strong performance in both static and dynamic world understanding, long-term consistency, and real-time generation efficiency, positioning it as a practical step toward interactive, memory-enabled world models.

Resources

Paper