WorldFlow preprint reports lower errors in predicting 3D motion

WorldFlow, a model that jointly predicts future video and three-dimensional motion, achieved a 54.2% reduction in average displacement error against the strongest external baseline on the SpatialCog evaluation set, its authors report in an arXiv preprint. Its average error on that set was 0.060 m. The model learns video evolution and motion trajectories together, using 1.44 million object-level training samples from simulated and real-world settings. It recorded the lowest average and final displacement errors among evaluated learning-based methods across three evaluation sets. These results establish benchmark performance rather than deployed robot capability.