Flora scales robotics data labeling at state of the art speed and cost
FLORA: An Open Model Suite for Scaling Robotics Data Engines
Robotics
Summary
Robots need lots of labeled data to learn and improve, but labeling can be slow and expensive. The authors present Flora, a powerful tool that labels many types of data for robot training more quickly and cheaply than other methods. Flora performs very well on multiple testing tasks and has been tested to label huge amounts of data every day. By sharing Flora openly, the authors hope others can build better robots and develop new ways for robots to learn from humans.
What this means in practice
- •For robotics data engineers: Label large-scale robotics datasets efficiently for training advanced robot models with multiple types of annotations like captions and rewards.
- •For automation operations teams: Scale real-time labeling and monitoring of physical robot performance data during large-scale deployments to improve system reliability.
Authors
Sicong Leng, Yunze Man, Zhiqi Li, Jing Wang, Yuhe Zhang, Ryan Punamiya, Siddharth Patki, Kaiyuan Zheng, Zi-ang Cao, Zhiding Yu, Sifei Liu, Danfei Xu, Ben Levin, Jan Kautz, Yuke Zhu, Johan Bjorck, Linxi "Jim" Fan
Abstract
The life-cycle of robotics foundation models crucially depends on data labels. Outcome reward labels are needed for monitoring real-world deployments. Process reward labels are needed during RL post-training. Timestamped captions are needed for filtering and balancing pretraining data. We present Flora, an industrial-quality data engine used for training GR00T models, that simultaneously covers these capabilities. Flora achieves SOTA scores on the captioning benchmark WGO-Bench and reward benchmarks RoboReward and Robometer. On internal benchmarks it is competitive with GPT-6 Astra while beating other frontier models. The model is substantially cheaper than API-based alternatives, roughly 125 times cheaper than GPT-6 Astra. Florais battle tested at scale, labeling 200,000 hours of data per day across 512 GB300s. By open-sourcing our data engine, we enable researchers to scale custom robotics datasets and explore emerging applications such as physical auto-research and human-to-robot transfer.