DriveLaW: Unifying Planning and Video Generation in a Latent Driving World

arXiv 25.12 2025 49.9 method, application

TLDR

DriveLaW unifies video generation and motion planning for autonomous driving by injecting latent representations from a world model into a diffusion planner.

Reasoning

The paper presents a novel unified paradigm with strong empirical results on video prediction and planning benchmarks. However, the abstract lacks details on limitations and the real-world scope of the benchmarks.

Read-first score

Read-first score 49.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 48.

Recency 6%

86.7

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

Topical relevance 29%

68.6

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 18%

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

Citation impact 18%

49.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.49707484

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 438.

Keyword Scores

world model

generative world model

video world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

interactive world model

Deep Analysis

Innovations

Unifying video generation and motion planning by directly injecting latent representation from video generator into planner
DriveLaW-Video: a powerful world model that generates high-fidelity forecasting with expressive latent representations
DriveLaW-Act: a diffusion planner that generates consistent and reliable trajectories from the latent of DriveLaW-Video
Three-stage progressive training strategy to optimize both components

Methodology

DriveLaW consists of two core components: DriveLaW-Video, a world model for high-fidelity video forecasting with expressive latent representations, and DriveLaW-Act, a diffusion planner that generates trajectories from the latent of DriveLaW-Video. Both components are optimized using a three-stage progressive training strategy. The model unifies planning and video generation by directly injecting the latent representation from the video generator into the planner.

Key Results

DriveLaW achieves state-of-the-art results on video prediction, surpassing the best-performing work by 33.3% in FID and 1.8% in FVD, and also sets a new record on the NAVSIM planning benchmark.

Resources

Paper