TableVerse: A Large-scale Tabletop Dataset with Real-world Grounded Layouts for Generalizable Manipulation
TableVerse turns unscripted internet images into 100,000 simulation-ready tabletop scenes for robot manipulation.
The paper describes a fully automated Real2Sim pipeline that reconstructs real-world tabletop layouts instead of hallucinating them from text. Its scenes are built with metric scale, dense clutter, authentic object arrangements, and checked mechanical stability. The dataset also pairs those environments with collision-free pick-and-place trajectories for training manipulation policies. HF Daily Papers' note
The paper describes a fully automated Real2Sim pipeline that reconstructs real-world tabletop layouts instead of hallucinating them from text. Its scenes are built with metric scale, dense clutter, authentic object arrangements, and checked mechanical stability. The dataset also pairs those environments with collision-free pick-and-place trajectories for training manipulation policies. HF Daily Papers' note
score 5