Capturing data…
Capturing data…
TUE, 27 OCT · 97 ITEMS
"FLORA: an open model suite for scaling robotics data engines" — arXiv cs.RO · AI & Technology
This arXiv paper from NVIDIA Research introduces , an open-source suite of three models—a temporal segmenter, a captioner, and a rewarder—that function together as a data engine for robotics. The suite automatically takes raw robot demonstration videos and produces timestamped action instructions and reward signals, dramatically lowering the cost of generating high-quality robot training data (the authors claim 125x cheaper than prior methods like Astra). FLORA is the actual data engine used to train NVIDIA's robot foundation model GR00T.
The claim originates directly from the arXiv preprint and NVIDIA's official research page, making it a primary-source report of what the paper states. The performance numbers (e.g., 125x cost reduction) are the authors' own reported results and have not been independently verified by third parties.
The paper was published on arXiv on October 26, 2026, so this is genuinely new material.