AI
WorldLine: Action-Driven Visual Simulation for Robotic Manipulation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 04:37:07 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 04:18:25 UTC.
Briefings
WorldLine is an action-driven visual simulator that decouples transferable dynamics learning from heterogeneous action grounding. It learns manipulation dynamics from over 10,000 hours of action-free robot videos and grounds them using over 2,000 hours of action trajectories across more than ten embodiments. The system uses an image-space action representation for shared control interfaces and multi-view training with relational regularization. It achieves strong visual quality and robot-motion agreement, improving robot-mask IoU by 0.1626 over baselines on failed trajectories. Performance on RoboTwin and AgiBot shows 74% mean accuracy, with rollouts improving task success by up to 21.4 percentage points over direct policy execution without prior training.
Key points
- WorldLine decouples transferable dynamics learning from heterogeneous action grounding using over 10,000 hours of action-free robot videos and 2,000 hours of action trajectories across ten embodiments.
- The system improves robot-mask IoU by 0.1626 over baselines on failed trajectories and achieves 74% mean accuracy on RoboTwin and AgiBot for trajectory prediction.
Context and limitations
Results are limited to held-out and out-of-domain settings; specific causal rollout mechanisms and failure-enriched training details are not fully detailed in the evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 04:37:07 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeic7qho62lpqc4wt27bynbogzbgj5rvr7p7qfyszcu3dxdpldgeykq