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

Kinematics-Aware Latent World Models for Data-Efficient Autonomous Driving

arXiv 26.3 2026 67.1 method, application

TLDR

A kinematics-aware latent world model improves sample efficiency and driving performance in autonomous driving simulation.

Reasoning

The paper presents a novel integration of kinematic information into RSSM-based world models, with geometry-aware supervision, showing clear improvements in simulation benchmarks. However, it lacks real-world validation and does not compare against recent large-scale world models.

Read-first score

Read-first score 67.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 8%
100

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

Methodology quality 25%
90

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

Topical relevance 42%
64.3

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 25%
38

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 150.

Keyword Scores

world model
10
model-based reinforcement learning world model
9
world dynamics prediction
8
generative world model
7
interactive world model
6
world simulator
3
video world model
2

Deep Analysis

Innovations

  • Incorporating vehicle kinematic information into the observation encoder to ground latent transitions in physically meaningful motion dynamics
  • Geometry-aware supervision to regularize the RSSM latent state to capture task-relevant spatial structure beyond pixel reconstruction

Methodology

The framework builds upon the Recurrent State-Space Model (RSSM) and integrates vehicle kinematic information into the observation encoder to enforce physically meaningful latent dynamics. Geometry-aware supervision is applied to regularize the latent state, enabling the model to capture spatial structure beyond pixel reconstruction. The approach is evaluated in a driving simulation benchmark against model-free and pixel-based world-model baselines.

Key Results

The proposed method achieves consistent gains in sample efficiency and driving performance over both model-free and pixel-based world-model baselines. Ablation studies further verify that the design enhances spatial representation quality within the latent space.

Tags