Source
Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage Buckets
The article details a workflow using Strands Agents, LeRobot, and Hugging Face Storage Buckets to create a continuous loop of recording robot demonstrations, training policies via streaming data, and deploying them to hardware.
Why it matters
This process reduces the time GPUs spend idle during downloads and lowers storage costs by only uploading changed bytes of large robotics datasets.
The details
- Xet-backed Storage Buckets use byte-level deduplication to minimize data transfer during uploads.
- Streaming directly from the Hub allows GPU training without full local dataset copies.
- The system supports simulation and physical robots such as the SO-101.
Show entities and relationshipsHide entities and relationships
In this article
Products
Companies
Countries
Key connections
AWS owns Strands Robots
AWS developed and open-sourced the Strands Robots SDK.
AWS owns Strands Agents
AWS provides the Strands Agents harness SDK.
Hugging Face owns Hugging Face Storage Buckets
Hugging Face developed and operates Hugging Face Storage Buckets.
Hugging Face owns Xet
Hugging Face provides the Xet storage backend for Storage Buckets.
NVIDIA develops Isaac Sim for robotics simulation.
NVIDIA created the Isaac-GR00T robot foundation model.
Show 20 more connectionsShow fewer connections
NVIDIA created the Cosmos 3 world foundation model.
Strands Robots uses lerobot
Strands Robots integrates the lerobot stack and dataset format as AgentTools.
Strands Robots uses Hugging Face Storage Buckets
Strands Robots uses Hugging Face Storage Buckets for syncing and streaming demonstration datasets.
Strands Robots uses Strands Agents
Strands Robots exposes robot tools that are composed into Strands agents.
Strands Robots uses Amazon Bedrock
Strands Robots can use Amazon Bedrock as a reasoning model provider for agents.
Strands Robots uses Isaac Sim
Strands Robots supports Isaac Sim as a simulation backend.
Strands Robots uses GR00T
Strands Robots supports training and fine-tuning on GR00T foundation models.
Strands Robots uses Cosmos 3
Strands Robots supports training and fine-tuning on Cosmos 3 foundation models.
Strands Robots uses SO-100
Strands Robots supports the SO-100 arm embodiment for simulation and real-world deployment.
lerobot supports leader-follower teleoperation and dataset collection on SO-100 and SO-101 robots.
Hugging Face Storage Buckets is built with Xet
Hugging Face Storage Buckets are backed by Xet for byte-level content-defined deduplication.
Hugging Face Storage Buckets is related to Hugging Face Hub
Hugging Face Storage Buckets operate alongside dataset repositories on the Hugging Face Hub.
Hugging Face Storage Buckets competes with Amazon S3
Hugging Face Storage Buckets provide a Hub-native alternative to Amazon S3 for streaming robotics datasets.
Amazon S3 competes with Hugging Face Storage Buckets
Amazon S3 serves as traditional cloud object storage compared against Hugging Face Storage Buckets.
Xet uses Content-Defined Chunking
Xet uses content-defined chunking to deduplicate data at the byte level.
Hugging Face Storage Buckets uses Content-Defined Chunking
Hugging Face Storage Buckets use content-defined chunking to reduce transfer sizes during dataset syncs.
Strands Robots uses Vision-Language-Action Models
Strands Robots supports training and inference of Vision-Language-Action models.
Strands Robots uses Action Chunking with Transformers
Strands Robots trains and deploys Action Chunking with Transformers policies.
lerobot uses Action Chunking with Transformers
lerobot includes ACTPolicy implementations for training and evaluation.
Hugging Face Storage Buckets is located in United States
Storage Buckets repositories are stored in US storage regions by default.
Related events
Announcement of Hugging Face Storage Buckets
AWS Introduces Streaming Data Loop Workflow for Strands Robots
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.