Sanping Zhou

Robotic world-action models do not need pretrained internet video backbones to scale: training native planning and dynamics modules from scratch across 30,000 hours unlocks strong zero-shot manipulation and drives a 17.7-point gain on an embodiment comprising under 2% of the data.

Policies trained with view augmentation from a novel feed-forward 3D Gaussian Splatting framework maintain robust execution under severe spatial perturbations where baselines fail.