World4Drive: End-to-End Autonomous Driving via Intention-aware Physical Latent World Model

ICCV 25 2025 60.4 method, application

TLDR

World4Drive uses vision foundation models to build a latent world model for perception-free, end-to-end autonomous driving planning.

Reasoning

The paper presents a novel framework that leverages vision foundation models to create a latent world model, enabling self-supervised planning without manual perception annotations. Strengths include state-of-the-art results on both open-loop and closed-loop benchmarks, while weaknesses are not explicitly discussed in the abstract, and reliance on simulated closed-loop evaluation may limit real-world generalization claims.

Read-first score

Read-first score 60.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 49.

Recency 8%

86.7

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

Topical relevance 42%

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=benchmark

Reproducibility 25%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 312.

Keyword Scores

world model

world dynamics prediction

generative world model

interactive world model

world simulator

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

End-to-end autonomous driving framework using vision foundation models to build latent world models without perception annotations
Intention-aware physical latent world model that generates and evaluates multi-modal planning trajectories
World model selector module to evaluate and select the best trajectory based on predicted future states

Methodology

World4Drive first extracts scene features including driving intention and world latent representations enriched with spatial-semantic priors from vision foundation models. It then generates multi-modal planning trajectories based on current scene features and driving intentions, predicts multiple intention-driven future states in latent space, and uses a world model selector module to evaluate and select the best trajectory. The framework achieves perception annotation-free, end-to-end planning through self-supervised alignment between actual future observations and predicted observations reconstructed from the latent space.

Key Results

World4Drive achieves state-of-the-art performance on open-loop nuScenes and closed-loop NavSim benchmarks without manual perception annotations, demonstrating an 18.1% relative reduction in L2 error, 46.7% lower collision rate, and 3.75× faster training convergence.

Resources

Paper