This work introduces SkyJEPA, a JEPA-style model designed for real-time quadrotor control that addresses the error amplification issues inherent in autoregressive long-horizon forecasting. The approach combines a latent dynamics model with a physics-inspired prober to map frozen latents to interpretable states, enabling physically grounded predictions.
- Combines a latent dynamics model with a novel physics-inspired prober for interpretable state mapping.
- Integrates the learned model with a sampling-based optimal control solution for real-time control on embedded hardware.
- Develops a structured pipeline for automated dataset generation to reduce reliance on expensive and unsafe real-world data collection.
- Demonstrates accurate prediction, robust zero-shot sim-to-real transfer, and strong generalization across diverse operating conditions in open-loop and outdoor closed-loop experiments.
The authors consider this important because it enables robust zero-shot sim-to-real transfer and accurate long-horizon prediction for agile aerial vehicles without requiring extensive real-world training data.