LeRobot Dataset Tools

Utilities for merging and validating LeRobot datasets.

This repository provides simple tools for:

• Merging multiple LeRobot datasets
• Verifying dataset integrity before training
• Inspecting dataset structure and statistics

---

Installation

Clone the repository:

git clone https://github.com/YOUR_USERNAME/lerobot-dataset-merge.git
cd lerobot-dataset-merge

Install dependencies:

pip install -r requirements.txt

---

Merge Datasets

You can merge multiple LeRobot datasets into a single dataset using the lerobot-edit-dataset tool.

Example:

HF_HUB_OFFLINE=1 lerobot-edit-dataset \
  --repo_id=/home/jacob/.cache/huggingface/lerobot/local/merged_datasettmt3cmra \
  --operation.type=merge \
  --operation.repo_ids='[
  "/home/jacob/.cache/huggingface/lerobot/local/demotmt50v3c",
  "/home/jacob/.cache/huggingface/lerobot/local/demotmt3cv3_2t"
  ]' \
  --new_repo_id=/home/jacob/.cache/huggingface/lerobot/local/merged_datasettmt3cmra \
  --push_to_hub=false

Explanation

Parameter	Description
repo_id	Target dataset directory
operation.type	Merge operation
operation.repo_ids	List of datasets to merge
new_repo_id	Output merged dataset
push_to_hub	Disable uploading to HuggingFace Hub

This command merges multiple local datasets into a single merged dataset.

---

Verify Dataset

After merging datasets, it is recommended to verify the dataset before training.

Run the verification script:

python verify.py

The script performs several checks:

1. Dataset basic information
2. Feature dimensions
3. Statistics validation
4. Data loading test

If the verification passes, the dataset is ready for training.

---

Dataset Structure

Typical LeRobot dataset structure:

dataset/
 ├── data
 ├── meta
 └── videos

---

Requirements

See requirements.txt.

Main dependencies:

• torch
• lerobot

---

License

MIT License