mimic-video: Video-Action Models for Robot Learning
Introduces Video-Action Models (VAMs) that replace standard VLMs in VLAs with a pretrained internet-scale video backbone (Cosmos-Predict2) combined with a flow-matching action decoder. Achieves approximately 10× better sample efficiency on real-world pick-and-place tasks.
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.