SparseWorld: Enhancing End-to-End Autonomous Driving via World Models with Sparse Scene Representation

arXiv 2026 61.8 method, application

TLDR

SparseWorld is a lightweight world model using sparse scene representation to predict future driving scenarios, improving end-to-end autonomous driving planning and reducing collision risk.

Reasoning

The paper introduces a novel sparse representation for world models, reducing computational cost while achieving state-of-the-art results on nuScenes and Bench2Drive benchmarks. However, the evaluation is limited to open-loop metrics on nuScenes and closed-loop on Bench2Drive, and potential limitations or failure cases are not discussed.

Read-first score

Read-first score 61.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 46.

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=baseline,benchmark,dataset,experiment,metric

Citation impact 18%

71.2

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.71235298

Topical relevance 29%

65.7

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

Reproducibility 18%

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

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: 419.

Keyword Scores

world model

world dynamics prediction

generative world model

interactive world model

model-based reinforcement learning world model

video world model

world simulator

Deep Analysis

Innovations

Sparse scene representation for world models to reduce computational cost and redundant information
Autoregressive rollout to forecast future map elements and surrounding agents
Sparse Dreamer with joint temporal and spatial attention for predicting future instances in latent space
Leveraging predicted futures to refine motion prediction and trajectory planning

Methodology

SparseWorld is a lightweight world model that first performs autoregressive rollout to forecast future map elements and surrounding agents, enabling the model to learn how driving scenarios evolve over time. It then uses these predicted futures to refine downstream motion prediction and trajectory planning via a Sparse Dreamer that anticipates future instances in latent space through joint temporal and spatial attention. The motion planner interacts with predicted future instances to capture more accurate motion patterns and generate safety-aware trajectories.

Key Results

SparseWorld achieves state-of-the-art performance on open-loop planning metrics of the nuScenes dataset with a collision rate of 0.05%, and substantially outperforms the baseline method in closed-loop planning metrics on the Bench2Drive benchmark.

Resources

Paper