Model tracks and generates robot tool-center-point trajectories from shared 4D scene representation
Researchers developed a model that uses a shared 4D scene representation to both track observed tool-center-point motion and generate future tool-center-point trajectories for robot manipulation.
A cross-embodiment augmentation pipeline converted existing RGB videos into demonstrations for alternative robot embodiments by retargeting recovered motion and synthesizing corresponding observations without recorded source-robot states.
The generation system achieved 69.19% mean success on one benchmark and surpassed established policy models on two real-world tasks, with joint training on source and augmented demonstrations raising target-robot success by 41.5% over augmented-only training.
Sources behind this report
- Hugging Face
Research · abstract · CC0 1.0
Source published
Read by the newsroom
Report history
No earlier revision dates are retained here. This does not establish that the report has never changed; read any dated updates and corrections.
This brief was made by AI and checked against its cited public sources. How the newsroom works