Our paper X-Sim has been accepted for an oral presentation at CoRL 2025! X-Sim introduces a real-to-sim-to-real framework that uses object motion from human videos to train robot policies without requiring any teleoperation data.