You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.


FR3 Pick-and-Place LeRobot Dataset

A real-robot demonstration dataset for pick-and-place with a Franka Research 3 (FR3) robot.

Each successful episode records a cube being picked from a table position and placed into a basket. The dataset is stored in LeRobot format and contains synchronized RGB camera observations, FR3 robot state, gripper state, joint-velocity actions, and gripper-closedness actions.

Task

Pick up the cube and place it in the basket.

Hardware

  • Franka Research 3 robot
  • Two Intel RealSense RGB cameras
  • Parallel gripper
  • Table workspace with a fixed basket

Data Format

The policy-facing action representation is pi0.5-DROID compatible:

  • 7 FR3 joint velocities
  • 1 gripper closedness value
  • 15 Hz control frequency

Successful recordings are validated before conversion to LeRobot. Failed or incomplete attempts are excluded from the released dataset.

Usage

from lerobot.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset("msbutt1/fr3-pick-place-lerobot")

See the accompanying FR3 robot-learning code repository for recording, conversion, training, and evaluation procedures.

Downloads last month
44

Models trained or fine-tuned on msbutt1/fr3-pick-place-lerobot