EgoWAM Framework for Human-to-Robot Manipulation Transfer
July 10, 2026
EgoWAM uses a co-training framework to improve human-to-robot task transfer by predicting world changes instead of just actions. By comparing Pixel, DINO, and 3D motion flow targets, the method aims to separate agent-invariant physical effects from non-transferable human morphology and camera motion.
HOW THIS AFFECTS YOU
●
builderThis provides a potential path for scaling robot manipulation using existing human video data.
●
researcherYou can use this co-training framework to decouple appearance from physical dynamics in egocentric datasets.