Back to events
Models
Score 74Heat 0

Nvidia open-sources Flora for cost-effective robot video labeling

Open SourceRoboticsOpen WeightsVisionMultimodalGPT-6 AstraNVIDIAFloraGR00TQwen3.5

Summary

Nvidia has released Flora, an open-source data engine built from fine-tuned Qwen models designed to annotate and score robot training videos at a significantly lower cost than commercial alternatives. The system combines action segmentation, captioning, and reward modeling to automate the preparation of robotics datasets.

Why it matters

Flora substantially reduces the cost of annotating robotics training data, offering an open alternative to proprietary systems and accelerating robot foundation model development.

Editorial score

2 reviews
7.5
Significance
7
Novelty
8.5
Credibility
7.5
Reach
6.5
Actionability