One loop, one bucket: how AWS's Strands Robots records, trains and deploys without redundant data transfers

Key points

• AWS released Strands Robots as an open-source toolkit (Apache 2.0 licence) that connects robot recording, model training and deployment inside a single AI agent loop.
• Hugging Face Storage Buckets, announced in March 2026, act as a mutable working layer where robot data lives between recording and training, avoiding repeated full-dataset downloads.
• The LeRobot dataset format, already used by more than 90,000 datasets from over 8,000 publishers, is the shared on-disk standard throughout the entire loop.
• The toolkit supports simulated runs on a laptop and physical robot arms such as the SO-101, switching between the two by changing a single keyword argument in code.
• Streaming from the Hub means each training run reads only the frames it needs, rather than copying the whole growing dataset to the GPU machine first.

Teaching a robot arm to pick up a cube takes more than one recording session. You collect demonstrations through the day, train a model on them, send that model back to the arm, watch it fail in a slightly new way, and record again. Do that loop by hand and you quickly notice the same bytes travelling back and forth over and over: upload the dataset, download the dataset to train, upload the new model, repeat.

AWS built Strands Robots, a free and open-source software toolkit, to close that loop automatically inside a single AI agent, a software program that can carry out multi-step tasks on its own without someone clicking through each stage.

What does the loop actually do?

Four stages share one backend, and the data never changes format between them. First, the agent records a demonstration: a continuous run of camera frames and joint-position readings from the robot arm, stored in LeRobot format, the open standard already used across tens of thousands of robot-learning datasets on Hugging Face, the AI model and dataset hosting platform.

Second, that dataset syncs into a Hugging Face Storage Bucket. A bucket is a storage space that lets software overwrite files in place, rather than saving every historical version the way a code repository does. Hugging Face announced this bucket type in March 2026. Because it sits inside the same Hugging Face workspace the developer already uses, there are no extra permissions to configure.

Third, training streams directly from the bucket, frame by frame, without downloading the whole dataset first. For a dataset that grows daily, this matters: a traditional approach copies everything to the training machine before the first gradient step runs, and that copy gets more expensive every day.

Fourth, the finished model checkpoint deploys back to the same robot object in code, switching from simulation to real hardware by changing one argument.

Who is this actually for?

Right now, the audience is robotics researchers and engineers who already use the LeRobot ecosystem. The simulation path runs on a laptop with no special hardware. The hardware path targets arms like the SO-101, a low-cost robot arm popular in academic labs, though the toolkit lists many other supported designs.

For anyone running a robot data-collection campaign, the practical benefit is less waiting and lower cloud-transfer costs. Each sync uploads only the bytes that changed since the last run, a technique called byte-level deduplication, rather than re-sending the full dataset.

The agent also handles decisions that otherwise require a human to check a screen: which episodes to keep, whether the scene has shifted enough to re-record, and which checkpoint replaces the one currently on the arm.

What happens next?

Strands Robots is available now on GitHub under the Apache 2.0 licence, meaning anyone can use or modify it freely. The companion notebook for this specific workflow is at examples/notebooks/05_streaming_data_loop.ipynb in the repository, first detailed by Hugging Face in their developer documentation.

The mock-policy default lets a developer run the full loop on a laptop before committing to GPU training time, which lowers the barrier to trying it. Real policy training still needs a GPU, and deploying to physical hardware still needs a calibrated arm.

The bigger picture is that the gap between "I recorded some robot data" and "my robot uses that data" keeps shrinking. Loops that once required a team coordinating uploads, training jobs and firmware flashes are starting to fit inside a few lines of code.