Strands Robots and Hugging Face Storage Buckets Streamline Robot Data Loop

AWS and Hugging Face have introduced a new workflow for robot learning using Strands Robots, an open-source SDK, and Hugging Face Storage Buckets. The system allows users to record demonstrations, train policies, and deploy them to physical robots from a single agent loop. Storage Buckets, announced in March 2026, are mutable, non-versioned, Xet-backed object storage that enables byte-level deduplication, reducing data transfer by about four times. The workflow includes recording demonstrations into a bucket, syncing only changed bytes, streaming datasets directly from the Hub for training, and deploying checkpoints to hardware with a single keyword argument change. The process uses LeRobot's dataset format, which is already used by over 90,000 datasets and models from more than 8,000 publishers. The agent loop can run continuously, collecting episodes throughout the day, training on the growing dataset, and deploying improvements. The system supports simulation and physical robots like the SO-101, and integrates with various policy providers including ACT, GR00T, and Cosmos 3. Security considerations include prompt injection risks and the need to separate credentials for writing and reading data. The full sample application is available on GitHub at strands-labs/robots.