OccDirector: Language-Guided Behavior and Interaction Generation in 4D Occupancy Space

arXiv 2026 62.1 method, system, application

TLDR

OccDirector generates 4D occupancy dynamics from natural language for autonomous driving simulation, achieving state-of-the-art instruction-following.

Reasoning

The paper introduces a novel framework for language-guided generation of 4D occupancy dynamics, with a new dataset and benchmark. Strengths include addressing the semantic-spatiotemporal gap and achieving high-quality generation. Weaknesses include a narrow focus on autonomous driving and potential scalability issues not discussed.

Read-first score

Read-first score 62.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 53.

Recency 6%

100

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

Methodology quality 18%

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

Topical relevance 29%

75.7

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

Citation impact 18%

64.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.64641027

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 366.

Keyword Scores

generative world model

world model

world simulator

interactive world model

world dynamics prediction

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Generates 4D occupancy dynamics conditioned solely on natural language without geometric priors
VLM-driven Spatio-Temporal MMDiT with history-prefix anchoring for long-horizon interaction consistency
OccInteract-85k dataset with multi-level language instructions from static layouts to multi-agent behaviors
Novel VLM-based evaluation benchmark for instruction-following

Methodology

OccDirector uses a VLM-driven Spatio-Temporal MMDiT architecture with a history-prefix anchoring strategy to generate 4D occupancy dynamics from natural language scripts. It is trained on the OccInteract-85k dataset, which provides multi-level language annotations, and evaluated using a novel VLM-based benchmark.

Key Results

OccDirector achieves state-of-the-art generation quality and unprecedented instruction-following capabilities, shifting the paradigm from appearance synthesis to language-driven behavior orchestration.

Resources

Paper