Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
WorldLine decouples scalable physical dynamics learning—trained on over 10,000 hours of action-free manipulation video—from heterogeneous embodiment grounding via a unified image-space action representation trained across ten robot platforms. By combining multi-view, failure-enriched training with robot-focused few-step distillation, it resolves the critical tension between generative video realism and precise physical action-following. The model improves robot-mask IoU by 0.1626 on failure cases, achieves 74% trajectory success prediction accuracy, and enables zero-shot visual rollouts that boost downstream policy success by up to 21.4 percentage points on RoboTwin.
Robot policies can now test and refine manipulation plans entirely within a learned visual simulator across ten distinct embodiments, gaining up to 21.4 percentage points in task success without any target-domain fine-tuning.
Real-world robot learning is constrained by the cost of collecting experience and evaluating candidate behaviors. Video generation models offer a scalable foundation for visual simulators that predict action outcomes before physical execution. Yet they often favor visual plausibility over accurate action following and coherent robot--object dynamics, while action-conditioned simulators depend on scarce, embodiment-specific data that are difficult to share across incompatible control spaces. We introduce WorldLine, an action-driven visual simulator that decouples transferable dynamics learning from heterogeneous action grounding. WorldLine learns manipulation dynamics from more than 10,000 hours of action-free robot videos and grounds them using over 2,000 hours of action trajectories across more than ten embodiments. An image-space action representation provides a shared control interface across embodiments, while multi-view and failure-enriched training with relational regularization improves interaction-sensitive prediction. Robot-focused few-step distillation enables efficient causal rollout while preserving action-critical motion. Across held-out and out-of-domain settings, WorldLine maintains strong visual quality and robot-motion agreement; on failed trajectories, it improves robot-mask IoU by 0.1626 over the strongest baseline. It predicts trajectory success with 74% mean accuracy across RoboTwin and AgiBot, one percentage point above the strongest baseline. Without RoboTwin training or adaptation, its rollouts improve task success by up to 21.4 percentage points over direct policy execution. Together, these capabilities make WorldLine a scalable and efficient visual simulator for policy evaluation and embodied planning. More results are available at https://zhengsh123.github.io/WorldLine/{project page}.