DexNDM: Closing the Reality Gap for Dexterous In-Hand Rotation

DexNDM bridges the sim-to-real gap for stable in-hand rotation of complex objects, enabling learning from biased real-world data without requiring any successful demonstrations.