SparseWorld: A Flexible, Adaptive, and Efficient 4D Occupancy World Model Powered by Sparse and Dynamic Queries

arXiv 25.10 2025 60.1 method

TLDR

SparseWorld proposes a 4D occupancy world model using sparse dynamic queries for flexible, adaptive, efficient perception, forecasting, and planning.

Reasoning

The paper introduces a novel approach with range-adaptive perception and state-conditioned forecasting, achieving state-of-the-art results across multiple tasks. However, it focuses narrowly on occupancy representation and does not address generative or interactive capabilities, limiting its scope.

Read-first score

Read-first score 60.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 33.

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=True; dataset=False; markers=github

Methodology quality 25%

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

Topical relevance 42%

47.1

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

Field roles

FrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 342.

Keyword Scores

world model

world dynamics prediction

video world model

generative world model

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Range-Adaptive Perception module that modulates learnable queries with ego vehicle states and temporal-spatial associations for extended-range perception
State-Conditioned Forecasting module that replaces classification-based forecasting with regression-guided formulation to align dynamic queries with 4D continuity
Temporal-Aware Self-Scheduling training strategy for smooth and efficient training

Methodology

SparseWorld is a 4D occupancy world model using sparse and dynamic queries. It employs a Range-Adaptive Perception module where learnable queries are modulated by ego vehicle states and enriched with temporal-spatial associations. A State-Conditioned Forecasting module uses regression-guided formulation instead of classification to capture scene dynamics. Training is facilitated by a Temporal-Aware Self-Scheduling strategy.

Key Results

SparseWorld achieves state-of-the-art performance across perception, forecasting, and planning tasks, with comprehensive visualizations and ablation studies validating its flexibility, adaptability, and efficiency.

Resources

Paper Code