Enhancing End-to-End Autonomous Driving with Latent World Model

arXiv 24.6 2024 52.7 method, application

TLDR

Proposes a self-supervised latent world model (LAW) that predicts future scene features to improve end-to-end autonomous driving, achieving SOTA on real-world and simulator benchmarks.

Reasoning

Strengths include a novel self-supervised approach that integrates seamlessly into existing frameworks and achieves state-of-the-art results on multiple benchmarks. Weaknesses are the limited methodological detail in the abstract and potential overclaiming without deeper analysis of limitations.

Read-first score

Read-first score 52.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 23.

Reproducibility 25%

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

Recency 8%

75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Methodology quality 25%

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

Topical relevance 42%

32.9

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

Field roles

Reproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 339.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Self-supervised learning approach using a latent world model (LAW) for end-to-end autonomous driving
Predicts future scene features based on current features and ego trajectories
Seamless integration into both perception-free and perception-based end-to-end driving frameworks

Methodology

The proposed LAW model employs a self-supervised learning task that predicts future scene features from current features and ego trajectories. It is designed to be integrated into end-to-end driving frameworks, including both perception-free and perception-based variants. The model is evaluated on real-world open-loop benchmarks (nuScenes, NAVSIM) and a simulator-based closed-loop benchmark (CARLA).

Key Results

LAW achieves state-of-the-art performance across multiple benchmarks: nuScenes, NAVSIM, and CARLA.

Resources

Paper Code