DreamTrue: action-faithful robot world model
An arXiv paper introduces DreamTrue, a multi-view, cross-embodiment robot world model for action-faithful and physically plausible video prediction. It renders action trajectories into image-space conditions with offline geometric calibration, adds counterfactual post-training to broaden interaction coverage, and uses a human-annotated defect dataset to train an embodied video reward model that guides reinforcement-learning post-training. On AgiBot, the authors report state-of-the-art action following, a drop in human-assessed interaction defect rate from 48.12% to 6.25%, and first place in the world model track of the AgiBot World Challenge 2026.
Action-faithful, physically plausible world models underpin robot learning and evaluation, so the reported defect-rate reduction and competition result are worth noting—pending replication and peer review.