NVIDIA open-sources Flora, a robotics data engine 125x cheaper than GPT-6 Astra
DrJimFan · x · 2026-10-21
Jim Fan's team released and open-sourced Flora, a model suite for robotics data engines that claims Astra-level performance at roughly 1/125 the cost, turning raw footage into rich training signals for robot learning.
- Capabilities: timestamped captions for curating egocentric/robot data, dense progress rewards for QC and RL post-training, and rubrics scoring for real-world benchmarking
- Scale: already in production, labeling 200K hours of video per day on 500 GPUs; it powers the data engine behind GR00T training
- Results: SOTA on the WGO-Bench captioning benchmark plus RoboReward and Robometer reward benchmarks; competitive with GPT-6 Astra on internal evals while beating other frontier models
- Weights are on Hugging Face, with an accompanying paper
Related event: NVIDIA Open-Sources Flora, a Robot Data Engine at 1/125 the Cost(3 posts)→
More from Embodied
- NVIDIA open-sources MotionBricks: 350K motion skills at 15,000 FPS, 2ms latency — aftahi_ai · 2026-10-21
- FLoTS-SLAM uses optical-flow residuals to hit 0.73 DM-IoU in dynamic monocular SLAM — zhenjun_zhao · 2026-10-21
- Plato ring launches at $199 to turn near-silent speech into AI commands — TinfoilTricorn · 2026-10-21
- Dream-1 controlled 31 unseen robot types across 53 countries within 24 hours — gharik · 2026-10-21
- Stanford's T2-z lets robots probe hidden physics with a poke, lifting success from 34% to 73% — stanfordnlp · 2026-10-21
- Snap Hires Robert Downey Jr. as Chief Customer Experience Officer to Sell Specs AR Glasses Against Meta — adariostrange · 2026-10-21