Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage Buckets

The blog post by Hugging Face introduces a data loop using Strands Agents, LeRobot, and Hugging Face Storage Buckets. It builds on the previous post about Strands Robots, an AWS open-source SDK (Apache 2.0) that exposes robot abstractions, simulation, and the LeRobot stack as AgentTools. The key idea is to record demonstrations, sync them to a Storage Bucket (a mutable, non-versioned, Xet-backed repository type), stream the dataset for training without full download, and deploy the trained checkpoint to hardware with a mode change. The example uses Robot("so100") to record a LeRobotDataset, sync_dataset_to_bucket to sync it, and stream_dataset to read it back. Storage Buckets are backed by Xet, which deduplicates uploads at byte level using content-defined chunking, reducing data transferred per upload by about four times. The post also covers recording on hardware with the lerobot-record CLI and notes that the same format works for both simulation and physical robots. It provides code snippets and prerequisites for running the loop.