ROS 2 → LeRobotDataset v3
Learn how units, timestamps, feature maps and license checks determine whether synthetic robot episodes are ready for conversion and a Hub dataset card.
Contribute to #93 ↗Open research · Student-focused · Independent project
Turn robotics questions into reproducible experiments.
Explore four proposed ways to connect ROS 2/Kubernetes with Hugging Face datasets, LeRobot policies, Jev comparison studies, and safe cloud–edge research. This demo is intentionally lightweight and does not need a robot, GPU or API key.
Research preview, not model inference: every scenario below is a preauthored synthetic example. No LeRobot policy was run, no Jev result was predicted, and no robot or Kubernetes system is controlled from this Space.
Four experiments to build
Choose a contribution
HF classifier versus Jev
Compare offline rules and a pinned open Hugging Face text classifier with optional Jev Choice using held-out synthetic DDS/QoS incident families.
Contribute to #94 ↗SmolVLA in LIBERO
Choose a genuinely compatible checkpoint and simulator before evaluating task success, action representation, camera mapping and repeatability.
Contribute to #95 ↗Remote policy inference
Study immutable Hub checkpoints, secure Zenoh-based LeRobot serving, bounded simulator actions and Kubernetes network-failure measurement.
Contribute to #96 ↗Interactive conceptual pipeline
Why the hard validation gate matters
Synthetic episode: task, timestamps and units are declared. Only an actual official LeRobotDataset conversion and reload could validate v3 compatibility.
Synthetic ROS 2 input
Deterministic gates
Optional HF artifact
Human review
Diagram illustrates a proposed workflow; neither branch actually runs a validator or uploads a dataset. On invalid input, a future validator would reject before any publication.
Evidence first, always
All published scenario outcomes are currently not measured. When a contributor has a real run, the project requests immutable model/data revisions, raw positive and failed attempts, code, seeds, measured metrics and independent review.
How to contribute reproducible results ↗