Robot Manipulation Data Recording System

5 posts · Nov 2024 – Jul 2026

What this trail covers

Hugging Face now supports Grabette, an open system for recording robot-manipulation data, enabling the conversion of Grabette recordings into LeRobot datasets. This facilitates the development and analysis of robot manipulation tasks. The system has been extended to allow users to use Strands Agents and LeRobot to deploy models from the Hugging Face Hub to robot hardware, enabling real-time control and interaction. The new v0.6.0 release of LeRobot introduces Robometer-4B, a 4B parameter model for on-device inference, and SmolVLM-Base, a 2B parameter image-text-to-text model, enabling efficient AI on embedded platforms. This post details the integration of these models and optimizations for on-device performance.

All Hugging Face trails

Jul 2026Grabette: an open system to record robot-manipulation dataThis post introduces Grabette, an open system for recording robot-manipulation data, and details its integration with LeRobot for dataset conversion. It highlights the ability to deploy models from the Hugging Face Hub to robot hardware using Strands Agents and LeRobot, enabling real-time control. The post also mentions the new v0.6.0 release of LeRobot and the Robometer-4B model.
May 2026From the Hugging Face Hub to robot hardware with Strands Agents and LeRobotThis post details the integration of the `allenai/MolmoAct2-SO100_101` model for robot manipulation. It showcases how to deploy models from the Hugging Face Hub to robot hardware using Strands Agents and LeRobot, enabling real-time control and interaction. The post highlights the use of the `allenai/MolmoAct2-SO100_101` model, a 5B parameter model, for robot manipulation tasks.
May 2026Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage BucketsThis post introduces the integration of Strands Agents with LeRobot and Hugging Face Storage Buckets, creating a unified platform for robot manipulation workflows. It highlights the ability to record data using Strands Agents, train models using LeRobot, and deploy them using Hugging Face Storage Buckets. The post also mentions the allenai/MolmoAct2-SO100_101 model, a 5B parameter model for robotics.
May 2026LeRobot v0.6.0: Imagine, Evaluate, ImproveThis post announces the v0.6.0 release of LeRobot, which introduces the Robometer-4B model. This release enhances the system's ability to imagine, evaluate, and improve robot behaviors. Key improvements include advanced simulation capabilities and the integration of the Robometer-4B model, a 4B parameter model specifically designed for robot manipulation tasks. The release also focuses on making it easier to deploy models from the Hugging Face Hub to robot hardware for real-time control and interaction.
Feb 2026LeRobot v0.5.0: Scaling Every DimensionThis post announces LeRobot v0.5.0, which introduces the fast-action-tokenizer for efficient action encoding in robotics. It also highlights the integration of Strands Agents and LeRobot for deploying Hugging Face Hub models to robot hardware, enabling real-time control. The release also includes Robometer-4B, a 4B parameter model for robot manipulation tasks.
Advertisement
Nov 2024Bringing Robotics AI to Embedded Platforms: Dataset Recording, VLA Fine‑Tuning, and On‑Device OptimizationsThis post details the integration of SmolVLM-Base (2B parameter image-text-to-text model) and Robometer-4B (4B parameter model) for on-device inference on embedded platforms. It discusses dataset recording using Grabette and conversion to LeRobot datasets, and VLA fine-tuning. Optimizations for on-device performance are highlighted, including quantization and efficient inference techniques for these models on resource-constrained hardware.