SkyJEPA: Learning Long-Horizon World Models for Zero-Shot Sim-to-Real Control of Quadrotors

arXiv 2026 53.8 method, application

TLDR

SkyJEPA combines latent dynamics with physics-inspired prober for long-horizon quadrotor control, achieving zero-shot sim-to-real transfer.

Reasoning

The paper introduces a novel JEPA-style model for quadrotor control, addressing long-horizon prediction errors via latent dynamics and a physics-inspired prober. Strengths include extensive real-world experiments and zero-shot transfer; weaknesses are domain specificity to quadrotors and lack of comparison to RL-based world models.

Read-first score

Read-first score 53.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 32.

Recency 6%

100

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

Citation impact 18%

89.4

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.89442086

Methodology quality 18%

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

Reproducibility 18%

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

Topical relevance 29%

45.7

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 330.

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

First application of JEPA-style model for real-time quadrotor control at high frequency
Novel physics-inspired prober that maps frozen latent representations to interpretable physical state
Structured pipeline for automated dataset generation to reduce reliance on real-world data
Zero-shot sim-to-real transfer demonstrated in outdoor closed-loop experiments

Methodology

The approach combines a latent dynamics model (JEPA) with a physics-inspired prober that decodes frozen latents into interpretable state, enabling physically grounded long-horizon prediction. The learned model is integrated with a sampling-based optimal control solution for real-time control on embedded hardware, and a structured pipeline automates dataset generation to avoid expensive real-world data collection.

Key Results

Open-loop and outdoor closed-loop experiments show accurate long-horizon prediction, robust zero-shot sim-to-real transfer, and strong generalization across diverse operating conditions.

Resources

Paper