skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

i4h-workflow-dataset-convert

Convert agentic HDF5 robotics recordings into LeRobot datasets with parquet, metadata, and videos.

3.2k|370|Updated Feb 25, 2026
Search Tags:#parquet#robotics#HDF5#dataset conversion#policy training#lerobot#isaac-for-healthcare
One-click install
npx skills add https://github.com/NVIDIA/skills --skill i4h-workflow-dataset-convert
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: i4h-workflow-dataset-convert
Source: https://github.com/NVIDIA/skills/tree/main/skills/i4h-workflow-dataset-convert
Command: npx skills add https://github.com/NVIDIA/skills --skill i4h-workflow-dataset-convert

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Robotics teams recording agentic HDF5 episodes with Isaac for Healthcare workflows need those recordings converted into LeRobot-format datasets before they can fine-tune policies such as GR00T, and doing this manually requires knowing the correct environment config, codec, and modality settings.

Core Features & Use Cases

  • HDF5 to LeRobot Conversion: Runs the i4h-workflows agentic dataset converter to produce parquet files, meta JSON, and per-episode videos under HF_LEROBOT_HOME.
  • Environment-Aware Defaults: Reads the env YAML (robot, task, cameras, dataset splits, modality) so the converter matches the recording that produced the HDF5.
  • Codec and Resolution Handling: Forces H.264 video encoding to avoid GR00T decord failures and normalizes mixed-resolution camera streams to the policy image size.
  • Use Case: After a teleoperation session produces a scissor_pick_and_place HDF5 recording, ask the agent to convert it and receive a ready-to-train LeRobot dataset with episode counts and output path reported.

Quick Start

Ask your agent to convert your latest scissor pick-and-place HDF5 recording into a LeRobot dataset for training.

Frequently Asked Questions about i4h-workflow-dataset-convert

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I convert an HDF5 robotics recording to a LeRobot dataset?▼

Set HDF5_PATH to your recording and run the i4h-workflows agentic dataset run.sh script with the same --env that produced the HDF5. The converter writes parquet files, meta JSON, and videos to HF_LEROBOT_HOME under your repo id.

What video codec should I use when converting to LeRobot for GR00T training?▼

Use --video-codec h264. The converter's default AV1 codec breaks GR00T's decord video reader at finetune time, so H.264 is required for training compatibility.

Does the converter handle cameras with different resolutions?▼

Yes. All camera streams are resized to the env YAML policy.image_size, normalizing mixed-resolution cameras such as head and overview cams. You can override the size with --image-size H W.

Why does the conversion fail with a modality config error?▼

This happens when the --env does not match the recording, so robot, task, camera, and modality defaults are wrong. G1 locomanip and assemble-trocar also require dataset.modality_template_path from the env YAML; scissor SO-ARM generates modality.json from YAML splits.

What prerequisites are needed before converting an HDF5 dataset?▼

The i4h workflow must be set up with its .venv via the i4h-workflow-setup skill, you need an existing HDF5 recording, and HF_LEROBOT_HOME must point to the desired output location.

Related Skills

Explore More Software Engineering Skills→
isaac-simisaac-sim
4.0k

urdf-mjcf-to-usd-conversion

Convert URDF and MJCF robot descriptions to USD assets for Isaac Sim and Isaac Lab.

Official
Advanced
bensonlee5bensonlee5

gr00t-n17-libero

Automate Franka Panda robot tasks with NVIDIA GR00T N1.7 VLA policy.

Community
Advanced
shushuznshushuzn
4

imaging-data-commons

Query NCI Imaging Data Commons metadata and download selected DICOM series.

Community
Intermediate
Hung-3008Hung-3008
1

vaex

Process and analyze out-of-core tabular datasets in CSV, HDF5, Arrow, and Parquet formats.

Community
Advanced