Challenge 2026 Dataset

This dataset provides training/evaluation data for the 2026 Global Humanoid Robot Challenge Simulation Competition.

Data Format: LeRobot v3.0

Directory Structure

Part_Sorting/                        - Desktop Fine-Grained Part Sorting
├── data/
├── meta/
├── videos/

Packing_Box/                         - Carton Sealing and Packing
├── data/
├── meta/
├── videos/

Foam_Inlaying/                       - Precise Part Assembly
├── data/
├── meta/
├── videos/

Conveyor_Sorting/                      - Conveyor Sorting
├── data/
├── meta/
├── videos/

Dataset Overview

• Desktop Fine-Grained Part Sorting

  • Part_Sorting: 3000 episodes

• Carton Sealing and Packing

  • Packing_Box: 1367 episodes

• Foam Inlaying

  • Foam_Inlaying: 539 episodes

• Covert Sorting

  • Conveyor_Sorting: 1064 episodes

Data Format

Each dataset contains multiple episodes stored in Parquet format:

LeRobot v3.0 (e.g., part_sorting_long_756_episode, Packing_Box, Foam_Inlaying, Conveyor_Sorting):

{dataset_name}/data/chunk-000/file-XXX.parquet

• Data filename: file-{file_index}.parquet

• Dataset ID format: {action}_{description} (e.g., part_sorting_long_756_episode)

Field definitions can be found in the main project at lerobot/common/policies/utils.py under observation/action definitions.

Hugging Face Download

Method 1: Hugging Face CLI

# Install HF CLI
powershell -ExecutionPolicy ByPass -c "irm https://hf.co/cli/install.ps1 | iex"

# Login
hf auth login

# Download
hf download UBTECH-Robotics/challenge2026_dataset

Method 2: Python

from huggingface_hub import login, snapshot_download

# Login
login()

# Download
snapshot_download(
    repo_id="UBTECH-Robotics/challenge2026_dataset",
    repo_type="dataset"
)

Method 3: Git + Git-XET

# Install git-xet
git xet install

# Clone (HTTPS)
git clone https://huggingface.co/UBTECH-Robotics/challenge2026_dataset

# Or use SSH
git clone git@hf.co:UBTECH-Robotics/challenge2026_dataset

Dataset Naming Convention