Generalization of World Models under Environmental Variability for Vision-based Quadrotor Navigation

arXiv 2026 59.3 method, application

TLDR

Study of DreamerV3 world model robustness to environmental variability in vision-based quadrotor navigation, with real-world deployment and open-loop imagination.

Reasoning

The paper provides a systematic evaluation of world model generalization under environmental variability, using both simulation and real-world quadrotor experiments, which is a strength. However, it is limited to a single model architecture (DreamerV3) and a specific task (quadrotor navigation), potentially reducing generalizability.

Read-first score

Read-first score 59.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 50.

Recency 6%

100

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

Citation impact 18%

83.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.83586785

Topical relevance 29%

71.4

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

Methodology quality 18%

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 399.

Keyword Scores

world model

generative world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

video world model

interactive world model

Deep Analysis

Innovations

Systematic study of world model robustness to environmental variability using vision-based quadrotor navigation as a testbed
Cross-environment validation spanning both Self-Supervised Learning (SSL) pretraining and Reinforcement Learning (RL) fine-tuning
Real-world deployment with an open-loop scenario where the model navigates entirely in imagination after only 2.5 seconds of sensory input
Identification of discrete latent size and training-sequence length as dominant factors governing world model quality

Methodology

The study uses DreamerV3-based world models trained under varying levels of environmental randomness. Models are evaluated via cross-environment validation across SSL pretraining and RL fine-tuning, and then deployed on a real quadrotor in unseen environments, including an open-loop test where the model receives 2.5s of real sensory input before navigating entirely in imagination over a 12m traverse.

Key Results

World model robustness during SSL pretraining is a strong predictor of sim-to-real transfer: every model that generalized well in cross-environment SSL validation deployed successfully in the real world (passing gaps as narrow as 0.67m), while the model that dominated simulation policy evaluation failed on the real platform. Discrete latent size and training-sequence length are identified as the dominant factors governing world model quality.

Resources

Paper