From 3c28b7143ac3fbb529534b37ce532a9ff3f0173a Mon Sep 17 00:00:00 2001 From: Bean Labs Date: Mon, 8 Jun 2026 14:31:13 -0400 Subject: [PATCH] Clarify Open-H checkpoint embodiment metadata --- README.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 1612b8c..029288a 100644 --- a/README.md +++ b/README.md @@ -6,7 +6,7 @@ [![Open-H Dataset](https://img.shields.io/badge/Dataset-Open--H-orange)](https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-Open-H-Embodiment) [![Python](https://img.shields.io/badge/Python-3.10+-blue.svg)](https://python.org) -A healthcare robotics variant of [GR00T N1.7](https://github.com/NVIDIA/Isaac-GR00T), post-trained on the [Open-H dataset](open_h/README.md) for multi-embodiment surgical and healthcare robot autonomy across 16 robot platforms and 34+ institutions. +A healthcare robotics variant of [GR00T N1.7](https://github.com/NVIDIA/Isaac-GR00T), post-trained on the [Open-H dataset](open_h/README.md) for multi-embodiment surgical and healthcare robot autonomy. The Open-H dataset spans 16 robot platforms and 34+ institutions.

GR00T-H Header @@ -16,6 +16,8 @@ A healthcare robotics variant of [GR00T N1.7](https://github.com/NVIDIA/Isaac-GR GR00T-H post-trains the GR00T N1.7 vision-language-action (VLA) foundation model on surgical robot data from multiple institutions and robot platforms simultaneously. Each institution records data differently: different robots, coordinate conventions, frame rates, camera setups, and state/action representations. GR00T-H addresses this by defining per-embodiment modality configs that convert each dataset into a common representation (`REL_XYZ_ROT6D` for EEF poses) while sharing the core VLA backbone. +The Open-H embodiment table documents dataset and modality-config coverage. Direct inference with a released Hugging Face checkpoint is limited to the embodiment tags included in that checkpoint's `processor_config.json` and `statistics.json`; use [`scripts/validate_hf_config_alignment.py`](scripts/validate_hf_config_alignment.py) to check the exact checkpoint metadata before running a new embodiment. + The primary additions over upstream Isaac-GR00T live in [`open_h/`](open_h/README.md): - Per-embodiment modality configs converting 16 healthcare robot datasets to a common action representation @@ -93,7 +95,7 @@ See [open_h/README.md](open_h/README.md) for a deeper dive on finetuning, multi- Open-H Dataset

-The [Open-H dataset](https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-Open-H-Embodiment) comprises 16 healthcare robot embodiments across 34+ institutions, stored in [LeRobot](https://github.com/huggingface/lerobot) format. See [open_h/embodiments/README.md](open_h/embodiments/README.md) for the full embodiment comparison table. +The [Open-H dataset](https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-Open-H-Embodiment) comprises 16 healthcare robot embodiments across 34+ institutions, stored in [LeRobot](https://github.com/huggingface/lerobot) format. See [open_h/embodiments/README.md](open_h/embodiments/README.md) for the full embodiment comparison table. That table is a dataset/config reference; a released model checkpoint may expose a smaller inference-ready set depending on its bundled processor and statistics metadata. ## Documentation