ProphetDWM: ProphetDWM: A Driving World Model for Rolling Out Future Actions and Videos

arXiv 25.5 2025 58.5 method, application

TLDR

ProphetDWM jointly predicts future driving videos and actions using an action module and diffusion model, achieving state-of-the-art on Nuscenes.

Reasoning

The paper introduces a novel end-to-end driving world model that addresses limitations of prior work by jointly learning action dynamics and video generation. Its strength lies in the joint prediction framework and strong empirical results on Nuscenes, but it is limited to a single dataset and lacks real-world deployment or diverse scenario evaluation.

Read-first score

Read-first score 58.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 8%

86.7

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

Topical relevance 42%

64.3

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=dataset,evaluation

Reproducibility 25%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 295.

Keyword Scores

world model

video world model

world dynamics prediction

interactive world model

generative world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Jointly predicts future videos and actions in an end-to-end driving world model
Action module to learn latent actions from present to future using action sequences and observations
Diffusion-model-based transition module to learn state distribution
Joint training that connects action dynamics and states for long-term future prediction

Methodology

ProphetDWM is an end-to-end driving world model with an action module that learns latent actions from the present to the future given action sequences and observations, and a diffusion-model-based transition module to learn state distribution. The model is jointly trained by learning latent actions from finite states and predicting both action and video. Evaluation is performed on the Nuscenes dataset for video generation and action prediction tasks, comparing against state-of-the-art methods.

Key Results

ProphetDWM achieves the best video consistency and best action prediction accuracy compared to state-of-the-art methods, while also enabling high-quality long-term video and action generation.

Resources

Paper