WorldLine: Action-Driven Visual Simulation for Robotic Manipulation
WorldLine uses robot video at scale to predict whether manipulation actions will work before a robot tries them.
The paper says the system separates broad dynamics learning from action grounding across different robot embodiments. It trains on more than 10,000 hours of action-free robot video and more than 2,000 hours of action trajectories from over ten embodiments. In tests, it preserved visual quality and robot-motion agreement, improved failed-trajectory robot-mask IoU by 0.1626 over the strongest baseline, and predicted trajectory success at 74% mean accuracy across RoboTwin and AgiBot. Without RoboTwin training or adaptation, its rollouts improved task success by up to 21.4 percentage points over direct policy execution. HF Daily Papers' note
The paper says the system separates broad dynamics learning from action grounding across different robot embodiments. It trains on more than 10,000 hours of action-free robot video and more than 2,000 hours of action trajectories from over ten embodiments. In tests, it preserved visual quality and robot-motion agreement, improved failed-trajectory robot-mask IoU by 0.1626 over the strongest baseline, and predicted trajectory success at 74% mean accuracy across RoboTwin and AgiBot. Without RoboTwin training or adaptation, its rollouts improved task success by up to 21.4 percentage points over direct policy execution. HF Daily Papers' note
score 5