Drive-JEPA: Video JEPA Meets Multimodal Trajectory Distillation for End-to-End Driving

arXiv 26.1 2026 32.5 method, application

TLDR

Drive-JEPA integrates Video JEPA with multimodal trajectory distillation for end-to-end autonomous driving, achieving state-of-the-art on NAVSIM.

Reasoning

The paper presents a novel combination of self-supervised video pretraining and trajectory distillation, with strong empirical results on a simulation benchmark. However, it lacks real-world validation and does not address generative or interactive world modeling.

Read-first score

Read-first score 32.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 18.

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,evaluation

Reproducibility 18%

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

Topical relevance 29%

25.7

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%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 156.

Keyword Scores

video world model

world model

world dynamics prediction

world simulator

generative world model

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Adaptation of Video Joint-Embedding Predictive Architecture (V-JEPA) for end-to-end driving, pretraining a ViT encoder on large-scale driving videos to produce predictive representations aligned with trajectory planning.
Introduction of a proposal-centric planner that distills diverse simulator-generated trajectories alongside human trajectories, with a momentum-aware selection mechanism to promote stable and safe behavior.

Methodology

Drive-JEPA first adapts V-JEPA for end-to-end driving by pretraining a ViT encoder on large-scale driving videos to learn predictive representations. It then uses a proposal-centric planner that distills both simulator-generated and human trajectories, employing a momentum-aware selection mechanism to choose stable and safe behaviors. The framework is evaluated on the NAVSIM benchmark in a perception-free setting, using a simple transformer-based decoder.

Key Results

The V-JEPA representation combined with a simple transformer-based decoder outperforms prior methods by 3 PDMS in the perception-free setting. The complete Drive-JEPA framework achieves 93.3 PDMS on NAVSIM v1 and 87.8 EPDMS on v2, setting a new state-of-the-art.

Limitations

Evaluation is limited to the NAVSIM simulation benchmark; real-world driving performance and generalization are not demonstrated.
The method relies on simulator-generated trajectories for distillation, which may not capture all real-world edge cases or driving scenarios.

Resources

Paper