| --- |
| license: apache-2.0 |
| task_categories: |
| - robotics |
| tags: |
| - manipulation |
| - vla |
| --- |
| |
| # Simulated Franka Pick-Cube Tactile Dataset in LeRobot Format |
| The dataset was generated using the [Robot Control Stack (RCS)](https://huggingface.co/papers/2509.14932). It is in LeRobot format and thus ready-to-train. The task is a simple "grasp the cuboid". |
| It is created by a [hardcoded trajectory planner](https://github.com/RobotControlStack/robot-control-stack/blob/master/examples/fr3/grasp_demo.py). |
| It contains tactile images from digit sensors attached to the fingers, rendered by [tacto](https://github.com/facebookresearch/tacto). |
|
|
| RCS is a flexible Gymnasium wrapper-based robot control interface made for robot learning and specifically Vision-Language-Action (VLA) models. |
| It unifies MuJoCo simulation and real world robot control with four supported robots: FR3/Panda, xArm7, UR5e and SO101. It ships with several pre-build apps including data collection via teleoperation and remote model inference. |
|
|
| Project page: https://robotcontrolstack.github.io/ |
| RCS Code: https://github.com/RobotControlStack/robot-control-stack |
|
|
| ## Dataset Structure |
|
|
| [meta/info.json](meta/info.json): |
| ```json |
| { |
| "codebase_version": "v3.0", |
| "robot_type": "fr3", |
| "total_episodes": 797, |
| "total_frames": 194468, |
| "total_tasks": 1, |
| "chunks_size": 1000, |
| "data_files_size_in_mb": 100, |
| "video_files_size_in_mb": 200, |
| "fps": 30, |
| "splits": { |
| "train": "0:797" |
| }, |
| "data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet", |
| "video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4", |
| "features": { |
| "observation.images.image": { |
| "dtype": "image", |
| "shape": [ |
| 256, |
| 256, |
| 3 |
| ], |
| "names": [ |
| "height", |
| "width", |
| "channel" |
| ], |
| "fps": 30.0 |
| }, |
| "observation.images.image2": { |
| "dtype": "image", |
| "shape": [ |
| 256, |
| 256, |
| 3 |
| ], |
| "names": [ |
| "height", |
| "width", |
| "channel" |
| ], |
| "fps": 30.0 |
| }, |
| "observation.images.tactile_left": { |
| "dtype": "image", |
| "shape": [ |
| 320, |
| 240, |
| 3 |
| ], |
| "names": [ |
| "height", |
| "width", |
| "channel" |
| ], |
| "fps": 30.0 |
| }, |
| "observation.images.tactile_right": { |
| "dtype": "image", |
| "shape": [ |
| 320, |
| 240, |
| 3 |
| ], |
| "names": [ |
| "height", |
| "width", |
| "channel" |
| ], |
| "fps": 30.0 |
| }, |
| "observation.state": { |
| "dtype": "float32", |
| "shape": [ |
| 15 |
| ], |
| "names": [ |
| "state" |
| ], |
| "fps": 30.0, |
| "description": "joints (7) + gripper (0 close, 1 open) (1) + tau_ext (7)" |
| }, |
| "action": { |
| "dtype": "float32", |
| "shape": [ |
| 8 |
| ], |
| "names": [ |
| "action" |
| ], |
| "fps": 30.0 |
| }, |
| "timestamp": { |
| "dtype": "float32", |
| "shape": [ |
| 1 |
| ], |
| "names": null |
| }, |
| "frame_index": { |
| "dtype": "int64", |
| "shape": [ |
| 1 |
| ], |
| "names": null |
| }, |
| "episode_index": { |
| "dtype": "int64", |
| "shape": [ |
| 1 |
| ], |
| "names": null |
| }, |
| "index": { |
| "dtype": "int64", |
| "shape": [ |
| 1 |
| ], |
| "names": null |
| }, |
| "task_index": { |
| "dtype": "int64", |
| "shape": [ |
| 1 |
| ], |
| "names": null |
| } |
| } |
| } |
| ``` |
|
|
|
|
| ## Citation |
| If you find this dataset or RCS useful for your work, please consider citing it: |
| ``` |
| @inproceedings{juelg2025robotcontrolstack, |
| title={{Robot Control Stack}: {A} Lean Ecosystem for Robot Learning at Scale}, |
| author={Tobias J{\"u}lg and Pierre Krack and Seongjin Bien and Yannik Blei and Khaled Gamal and Ken Nakahara and Johannes Hechtl and Roberto Calandra and Wolfram Burgard and Florian Walter}, |
| year={2025}, |
| booktitle={Proc.~of the IEEE Int.~Conf.~on Robotics \& Automation (ICRA)}, |
| note={Accepted for publication.} |
| } |
| ``` |
