# KETI Humanoid HRI Safety Dataset: Simulated Unitree H1 Warehouse Collisions and Near-Misses Labeled against ISO/TS 15066

## Facts

- Provider: Korea Electronics Technology Institute (KETI)
- License: CC-BY-4.0 (https://huggingface.co/datasets/KETI-IRRC/keti-humanoid-hri-safety)
- Size: 171 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/KETI-IRRC/keti-humanoid-hri-safety
- Access: public, readable without an account
- Category: Robotics
- Source: https://github.com/safety-cybersecurity/keti-humanoid-hri-safety
- Page: https://scigantic.com/archive/8843e2f1-6848-53d7-9da6-1cb7e46dac6b
- Keywords: Robotics, Parquet, Hugging Face

## About

A public dataset for evaluating the safety of humanoid robots working around people (human-robot interaction, HRI), from the Korea Electronics Technology Institute. In a simulated logistics warehouse in NVIDIA Isaac Sim, a Unitree H1 humanoid carrying a 1 to 3 kg box with the published Isaac-Velocity-Flat-H1-v0 walking policy meets a jointed human dummy in three scenarios where it cannot see the person in time: a corner hidden by a rack, a person stepping into the aisle, and the robot turning round with a box. The baseline robot does not react, so the data hold failures, near-misses and boundary cases. Each scenario has 99 conditions (50 random and a 7x7 grid that forms a failure map), 297 episodes and 247,181 steps at 50 fps, with seven 640x480 RGB cameras in AV1 video. Labels are computed, not annotated: contact forces on twelve body regions are compared with ISO/TS 15066 Annex A limits and separation distance with the protective separation distance, giving outcome (fail, near_miss, nominal), severity S0-S3 and the triggered rules. Format is LeRobotDataset v3.0 with a safety extension, plus the USD assets; about 20 GB (measured 2026-11-21). Data CC BY 4.0, Unitree H1 files BSD-3-Clause, per the card and NOTICE.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("KETI-IRRC/keti-humanoid-hri-safety")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/chunk-000/file-000.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/8843e2f1-6848-53d7-9da6-1cb7e46dac6b
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/8843e2f1-6848-53d7-9da6-1cb7e46dac6b
- Over MCP: `get_schema_card` with id `8843e2f1-6848-53d7-9da6-1cb7e46dac6b` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Robotics datasets](https://scigantic.com/topics/robotics)
