---
title: "Lerobot SO-101 Manipulations Dataset"
description: "Lerobot SO-101 dataset is a compact, high-quality set of 20 teleoperated robotic arm recordings (~30 FPS) captured on the calibrated robot studio SO-101 arm using…"
url: "https://unidata.pro/datasets/lerobot-so-101-manipulations/"
date_modified: "2026-05-19T14:53:14+03:00"
language: "en-US"
---
Lerobot SO-101 dataset is a compact, high-quality set of 20 teleoperated robotic arm recordings (~30 FPS) captured on the calibrated robot studio SO-101 arm using multi-view cameras (front, wrist, top) in the Lerobot framework. Designed for imitation learning and model training on the Lerobot platform, it demonstrates pick-and-place and basic manipulation tasks with structured LeRobot v2.1 annotations.

This dataset is intended as a limited reference set and contains 20 episodes; however, full custom data collection in Lerobot format can be provided on request.

## Dataset Structure

### The Numbers Section

**Numbered list:**

- **Number:** 20 — **Text:** Episodes
- **Number:** ~30 — **Text:** FPS

### Dataset Information

**Table with data:**

| Characteristic | Data |
| --- | --- |
| Description | Robotic manipulation demos (e.g., pick-and-place) for imitation learning |
| Data types | Video |
| Tasks | Pick-and-place, basic manipulation |
| Number of episodes | 20 |
| Episode length | Variable (~30-60s per episode) |
| Labeling | Teleoperated human demonstrations; meta.json with stats |
| Robot | SO-101 (5-DOF arm + gripper) |

**Media Slider:**

- **Image in the slider:** ![robot dataset](https://unidata.pro/wp-content/uploads/2026/01/3d-rendering-metal-cyborg-arm-isolated-white-scaled.webp)
- **Image in the slider:** ![manipulations dataset](https://unidata.pro/wp-content/uploads/2026/01/86z_robotic_arm_in_one_line_blue_sh-scaled.webp)

**Link to the sample:** [Download sample](https://drive.google.com/drive/folders/16Xkvp38vWbANeh7eHbRP3I8Q0vgeO2lU)

### Technical Specifications

**Table with data:**

| Characteristic | Data |
| --- | --- |
| Format | LeRobot v2.1 (Parquet + MP4) |
| FPS | 30 |
| Cameras | front, wrist, top |
| State dim | 7 (joints + gripper) |
| Action dim | 7 |

**Source and data collection methodology:** Source and collection methodology: Data recorded via teleoperation on SO-101 robot using Lerobot-record

### Dataset Use Cases - Slider

**Industry Cards:**

- **Industry:** Robotics Research — **Title:** Imitation Learning for Manipulation Tasks — **Text:** This dataset supports imitation learning by providing teleoperated demonstrations recorded on a calibrated SO-101 arm. Researchers can use these recordings to train and evaluate models for pick-and-place and basic manipulation, leveraging synchronized video, joint states, and actions within the lerobot framework.
- **Industry:** Industrial Automation — **Title:** Rapid Prototyping of Robotic Arms — **Text:** For industrial teams working with robotic arms, the so-101 dataset offers a practical baseline for testing control logic and model training pipelines. The structured recordings from front, wrist, and top cameras help validate robotic workflows before scaling, reducing setup time for robot studio SO-101 arm deployments.
- **Industry:** Education and Training — **Title:** Hands-On Learning with Open-Source Robotics — **Text:** This dataset is well suited for teaching robotics concepts using the open-source lerobot platform. Students can explore teleoperation, calibration, and data formats while analyzing real SO-101 camera feeds and state-action logs, gaining practical experience with robotic systems and model evaluation.
- **Industry:** Applied AI Development — **Title:** Custom Data Collection for Model Training — **Text:** Although limited in size, this dataset demonstrates how data can be recorded using start teleoperation on so arm 101 setups. It serves as a reference for teams planning custom data collection, enabling tailored training models and pretrained model fine-tuning for specific robotic tasks and environments.

### Fact

**FAQs Heading:** FAQs

**List of Questions:**

- **Question:** What types of annotations are provided with this dataset? — **Answer:** Annotations are derived from teleoperated human demonstrations. A meta.json file provides structured statistics for each episode, supporting reproducible model training and evaluation.
- **Question:** What are the sources of data for Unidata robotics datasets? — **Answer:** The dataset was recorded using the open-source Lerobot platform. All episodes were captured via teleoperated human demonstrations on the SO-101 robotic arm.
- **Question:** What data formats and technical characteristics are included? — **Answer:** Data is provided in LeRobot v2.1 format using Parquet files and MP4 videos at 30 FPS. Recordings include front, wrist, and top camera views with 7-dimensional state and action spaces.
- **Question:** How are Unidata datasets licensed? — **Answer:** Unidata follows a dual-licensing model. Free samples are available for testing, while the complete dataset is provided exclusively through purchase.
- **Question:** Do Unidata datasets comply with GDPR and data privacy regulations? — **Answer:** Yes. All datasets comply with GDPR and applicable data protection laws. Data is collected from legally permissible sources to ensure ethical usage.
- **Question:** How are Unidata datasets stored and secured? — **Answer:** All datasets are securely stored on AWS cloud infrastructure. Storage and data management practices follow ISO 27001 and ISO 27701 standards for security and privacy.
- **Question:** How long does it take to receive the dataset after purchase? — **Answer:** After request submission, Unidata reviews the details and completes documentation. Delivery typically occurs within 3–10 days after signing and payment.
- **Question:** What manipulation tasks are included in the dataset? — **Answer:** It contains 20 teleoperated episodes demonstrating pick-and-place and other basic manipulation tasks using the SO-101 robotic arm. Each recording lasts approximately 30–60 seconds, capturing realistic robot control sequences for learning manipulation policies.
- **Question:** What makes this Unidata dataset valuable for robotics research? — **Answer:** Unlike generic robot video collections, this dataset combines teleoperated demonstrations, multi-view synchronized cameras, structured LeRobot v2.1 annotations, and calibrated SO-101 robot state-action data. It provides an immediately usable benchmark for imitation learning, robot manipulation, and Physical AI research.

## List of Parameters

- **Title:** Tasks — **Description:** Pick-and-place, manipulation
- **Title:** Format — **Description:** LeRobot v2.1 (Parquet + MP4)
- **Title:** Sensor data — **Description:** 3 camera views, 7-axis state and action space
- **Title:** Episodes — **Description:** 20
- **Title:** Training — **Description:** Compatible with imitation learning pipelines

[Full list of this site's AI-readable pages](https://unidata.pro/llms.txt)