Nvidia open-sources Flora for robot video labeling
Nvidia has released Flora, an open-source data engine designed to label robot training videos. The system uses three fine-tuned Qwen3.5 models to generate action descriptions, progress ratings, and success judgments. Nvidia estimates that labeling one hour of video costs 72 cents with Flora, compared to roughly $88.50 using GPT-6 Astra. These cost figures are projections based on an assumed GPU…
Key points
- Nvidia open-sources Flora, using three fine-tuned Qwen3.5 models for robot video labeling.
- Nvidia estimates Flora costs 72 cents per hour of video, versus $88.50 for GPT-6 Astra.
- Flora detects mid-episode errors better than competitors, with a 27.8% lower error rate than Astra.
In tests, Flora outperformed GPT-6 Astra in detecting action boundaries and identifying mid-episode errors. On RoboRewardBench, Flora’s error rate was 27.8 percent lower than Astra’s. However, Astra often provided more precise descriptions and led in rubric-based scoring. Nvidia notes that Flora is already running in the training of its GR00T models, processing up to 200,000 hours of video per day across 512 GB300 GPUs.
The release positions Flora as an open alternative to commercial services like Scale AI’s Physical AI Data Engine. It aims to help researchers scale robotics datasets and test approaches like transferring human demonstrations to robots. This move supports Nvidia’s broader strategy to cover the entire robotics value chain, from hardware to simulation and AI models, while driving demand for its chips.
Model page: GPT-6 Astra →
Nvidia open-sources Flora to make robot video labeling 125x cheaper than GPT-6 Astra
The Decoder · 25 October 2026
Loading the full article…
This text was published by The Decoder and written by Jonathan Kemper. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Anthropic co-founder Jack Clark says self-improving AI could arrive by 2027 · 1 src
- OpenAI keeps frontier models offline after September DNS escape and red-teaming finds · 22 src
- Microsoft Open-Sources Agensh, a Multi-Agent harness for coding · 1 src
- MarkTechPost publishes tutorial on running Underdog Saluki 27B quantized model · 1 src
- OpenAI agents study Parkinson-like disease data in 32 minutes · 1 src
Comments
via GitHub Discussions