SkyJEPA: Learning Long-Horizon World Models for Zero-Shot Sim-to-Real Control of Quadrotors
Pratyaksh Rao, Wancong Zhang, Randall Balestriero, Yann LeCun, Giuseppe Loiano
THE PROBLEM
This paper focuses on world models. This paper shows how to train quadrotor controllers in Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. without manual data collection by using a latent-space Modern Robot LearningWorld modelA model that predicts how the world will change after actions. (JEPA) that predicts accurately over long horizons without error accumulation. The result is real quadrotors flying aggressively outdoors with zero Modern Robot LearningFine-tuningTaking a pretrained model and adapting it to a specific robot or task. after sim Robot LearningTrainingThe process of fitting a model using data or experience., solving a major bottleneck in aerial robotics where real-world flight data is expensive and dangerous to collect. Read the paper by tracking the Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. definition, the Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or data assumptions, and the evidence that supports the claimed improvement.
HOW IT WORKS
Task framing
Core method
Data and supervision
Evaluation evidence
KEY RESULTS
This paper shows how to train quadrotor controllers in Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. without manual data collection by using a latent-space Modern Robot LearningWorld modelA model that predicts how the world will change after actions. (JEPA) that predicts accurately over long horizons without error accumulation. The result is real quadrotors flying aggressively outdoors with zero Modern Robot LearningFine-tuningTaking a pretrained model and adapting it to a specific robot or task. after sim Robot LearningTrainingThe process of fitting a model using data or experience., solving a major bottleneck in aerial robotics where real-world flight data is expensive and dangerous to collect.
WHY DEVELOPERS SHOULD CARE
This paper shows how to train quadrotor controllers in Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. without manual data collection by using a latent-space Modern Robot LearningWorld modelA model that predicts how the world will change after actions. (JEPA) that predicts accurately over long horizons without error accumulation. The result is real quadrotors flying aggressively outdoors with zero Modern Robot LearningFine-tuningTaking a pretrained model and adapting it to a specific robot or task. after sim Robot LearningTrainingThe process of fitting a model using data or experience., solving a major bottleneck in aerial robotics where real-world flight data is expensive and dangerous to collect.
LIMITATIONS
The main limitation to check is whether the claimed behavior holds outside the paper's reported setup. That means testing across different Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. embodiments, scenes, objects, and data distributions.
WHAT COMES NEXT
The practical next step is independent reproduction with clear baselines, ablations, and stress tests. For a developer, the useful follow-up is to map the paper's world models assumptions onto a concrete Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. stack, then test the smallest version of the method that could run end to end.