⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

Enhance Sample Efficiency and Robustness of End-to-end Urban Autonomous Driving via Semantic Masked World Model

TITS 2026 62.4 method, application

TLDR

Proposes SEM2, a semantic masked recurrent world model with multi-source data sampler to improve sample efficiency and robustness in end-to-end urban autonomous driving.

Reasoning

The paper introduces a novel semantic filter and data balancing technique to address task-irrelevant latent states and data imbalance, which are clear strengths. However, it only evaluates on the CARLA simulator without real-world experiments, limiting generalizability.

Read-first score

Read-first score 62.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 38.

Recency 6%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%
93.5

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.93501976

Citation velocity 12%
69.3

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=29.00

Methodology quality 18%
60

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

Topical relevance 29%
54.3

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

Reproducibility 18%
30

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FoundationFrontier

Rank sensitivity

Stability: volatile; rank range: 508.

Keyword Scores

world model
10
model-based reinforcement learning world model
8
video world model
7
world dynamics prediction
6
interactive world model
5
world simulator
2
generative world model
0

Deep Analysis

Innovations

  • Semantic filter to extract key driving-relevant features from latent states
  • Multi-source data sampler that aggregates common data and multiple corner case data in a single batch to balance data distribution

Methodology

The paper proposes SEM2, a semantic masked recurrent world model that uses a semantic filter to extract task-relevant features from latent representations, and a multi-source data sampler to balance training data distribution by mixing common and corner case data. The model is trained end-to-end on the CARLA simulator.

Key Results

Extensive experiments on CARLA show that SEM2 outperforms state-of-the-art approaches in sample efficiency and robustness to input perturbations.

Resources

Paper

Tags