Structured data model

GE Act 2.0

GE-Act 2.0 is a world-action model for robotic manipulation from AGIBOT Research, combining a control-oriented autoencoder, a single-step visual planner, and an inverse dynamics model. Trained on action-free video plus action-labeled data, it predicts future visual states and generates matching robot actions, enabling zero-shot manipulation across 100 tasks and cross-embodiment transfer. Scaling co-training data from 300 to 30,000 hours raised task success from 17.1% to 44.1%.