OCCVAR: Scalable 4D Occupancy Prediction via Next-Scale Prediction

OpenReview 2026 51.5 method, application

TLDR

OCCVAR proposes a generative occupancy world model using next-scale prediction for efficient, high-quality 4D occupancy forecasting.

Reasoning

The paper introduces a novel spatial-temporal transformer with temporal next-scale prediction to address inefficiency and temporal degradation in autoregressive occupancy models. Strengths include fast inference and long-time generation; weaknesses are the lack of explicit real-world dataset details in the abstract.

Read-first score

Read-first score 51.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 39.

Recency 8%

100

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

Topical relevance 42%

55.7

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: 515.

Keyword Scores

generative world model

world model

world dynamics prediction

world simulator

video world model

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Next-scale prediction for 4D occupancy scenes, enabling coarse-to-fine generation
Multi-scale scene tokenizer to capture hierarchical 3D geometry
Incorporation of ego movement into tokenized occupancy sequence for controllable scene generation

Methodology

OCCVAR uses a spatial-temporal transformer with temporal next-scale prediction to generate 4D occupancy scenes from coarse to fine scales. It tokenizes the occupancy sequence and incorporates ego movement before the tokens to model dynamic evolution. A multi-scale scene tokenizer captures hierarchical 3D geometry information.

Key Results

OCCVAR achieves high-quality occupancy reconstruction, long-time generation, and fast inference speed compared to prior works.

Resources

Paper