Nvidia has open-sourced Flora, a new data engine designed to label robot video footage at a fraction of the cost and with higher quality than existing solutions. According to The Decoder, Flora can label one hour of robot video for just 72 cents, compared to $88 per hour using GPT-6 Astra, making it approximately 125 times cheaper.
The Decoder reports that Flora is powered by three fine-tuned Qwen3.5 models. While Flora outperforms competitors in catching errors in robot actions, it occasionally lacks descriptive accuracy. Nvidia is positioning Flora as a strategic tool to gain control over the entire robotics value chain.
This development could be significant for Japanese markets, where robotics and automation continue to be key growth areas, potentially lowering costs and improving quality in robot training and deployment.