Researchers have introduced DreamTrue, a multi-view, cross-embodiment robot world model designed for action-faithful and physically plausible video prediction. The work addresses two key obstacles in training such models on existing robot datasets: imprecise calibration that can impair action following, and limited coverage of unsuccessful interactions that can bias predictions toward successful outcomes.
To improve action following across embodiments, the team renders action trajectories into image-space conditions and applies offline geometric calibration to align these conditions with target videos. To broaden interaction coverage, they introduce counterfactual post-training, which modifies recorded action trajectories and generates future videos under a wider range of actions and contact configurations.
Because paired ground-truth futures are unavailable, the researchers built a human-annotated video dataset covering robot, object, and interaction defects, and used it to train an embodied video reward model. Its scores guide reinforcement-learning post-training toward more physically plausible interaction outcomes.
On AgiBot, DreamTrue achieves state-of-the-art action following while reducing the human-assessed interaction defect rate from 48.12% to 6.25%. The model also ranks first in the world model track of the AgiBot World Challenge 2026. The project page is available at https://brave-eai.github.io/DreamTrue.