DreamTrue
Dataset Analysis
DreamTrue is a multi-view, cross-embodiment robot world model for action-faithful video prediction, using geometric calibration and counterfactual post-training with an embodied reward model.
Provenance
Collected from papers.
Derived from paper: DreamTrue: Action-Faithful Robot World Model with Counterfactual Post-Training