A Hugging Face post from Amazon describes a more continuous workflow for collecting robot demonstrations, training policies, and deploying them back to hardware. The setup combines Strands Robots, LeRobot, and Hugging Face Storage Buckets so the same data loop can run without repeatedly copying full datasets around.
The problem is practical. Once teams collect robot demonstrations throughout the day, each new training run can become expensive if it downloads the growing dataset to GPUs, writes checkpoints elsewhere, and then repeats that cycle for the next batch. The post frames Storage Buckets as a way to keep the LeRobot-format dataset accessible in the same place across recording, training, and deployment.
Strands Robots is an open source SDK from AWS that exposes robot abstractions, simulation, and LeRobot tools through agent-compatible interfaces. The earlier workflow covered recording demonstrations and deploying policies; this update focuses on making the loop sustainable as data grows.
For robotics teams, the value is less about a new model and more about infrastructure. A cleaner data path can make iterative robot learning easier to run every day, though hardware quality and dataset design still determine how useful the trained policy becomes.