embodiment-gap

This paper presents HuRo, a pipeline for robotizing human videos to create scalable VLA pretraining data, showing significant improvements in task completion and robustness on real-world manipulation tasks.

The article discusses the embodiment gap in robotics and how Bonafide Robotics is addressing it by providing industrial robot fleets on a subscription basis.

Dyna Robotics introduces Dyna-2, a world-action model pre-trained on one million hours of human video, revealing scaling laws and suggesting the embodiment gap is an adaptation problem rather than a knowledge problem.

This paper identifies an embodiment gap in humanoid co-speech motion generation caused by human-centric pipelines, and proposes PhysDrift, an embodiment-aware framework that directly predicts executable humanoid joint trajectories from speech, improving speech-motion alignment and physical plausibility.