josabb/phuma_h1_2_pdfhr

---
license: apache-2.0
task_categories:
- reinforcement-learning
- robotics
tags:
- code
pretty_name: Unitree G1 Velocity Training
size_categories:
- 10M<n<100M
---

# PHUMA H1_2 PDF-HR Dataset

This dataset contains processed and prepared poses of the H1_2 robot (PHUMA) for training the PDF-HR model
(Pose Distance Field for Human-Robot). It is derived from the original PHUMA H1_2 dataset and specifically
adapted for the experiments in the PDF-HR paper.

## Dataset Contents

The dataset includes:

- `positive_poses_all.npy`: positive poses with distance 0
- `near_poses.npy` and `near_distances.npy`: poses near the manifold, with corresponding distances
- `interp_poses.npy` and `interp_distances.npy`: interpolated poses between others, with corresponding distances

All arrays are in **NumPy `.npy` format**.

## Robot

- **Robot model:** H1_2 (PHUMA)
- **Degrees of Freedom:** 27 active joints (revolute)
- **Joint order:** as defined in the PDF-HR paper (matches URDF joint hierarchy)