Flora open robotics data engine reports SOTA caption and reward scores
Original title: FLORA: An Open Model Suite for Scaling Robotics Data Engines
Authors say Flora labels 200,000 hours a day and is about 125 times cheaper than GPT-6 Astra.
Flora is an industrial data engine used to train GR00T models, producing outcome and process reward labels plus timestamped captions. The authors report state-of-the-art scores on WGO-Bench, RoboReward and Robometer, and say it is competitive with GPT-6 Astra on internal benchmarks while beating other frontier models. They report labeling 200,000 hours of data per day across 512 GB300s and are open-sourcing the engine for custom robotics datasets.
Unconfirmed: Cost, throughput and GPT-6 Astra comparisons are figures reported by the authors.
- Flora covers outcome rewards, process rewards and timestamped captions.
- Authors report SOTA on WGO-Bench, RoboReward and Robometer.
- They say Flora is roughly 125 times cheaper than GPT-6 Astra.
- Claimed throughput is 200,000 hours of data per day on 512 GB300s.
From the original article
1 min read at huggingface.co
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.
By Sicong Leng, Yunze Man, Zhiqi Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.