RynnWorld-Teleop: Digital Teleoperation via Generative World Models

Haoyu Zhao · hf · 2026-07-08

This study proposes replacing physical robot interactions with action-conditioned generative world models for digital teleoperation scenarios. By generating diverse training data, it enables efficient zero-shot Sim2Real transfer and significantly improves real-world execution performance. This solution lowers the cost of collecting real-world robot operation data, helping to accelerate the generalization training of robotic policies.

Original post →

More from Embodied

Embodied channel →