AtomBench FrankaPanda Dataset

A robot manipulation dataset collected on a Franka Panda robot arm, designed for evaluating robotic learning policies. The dataset contains 15 tasks (8 manipulation tasks + 7 instruction-following tasks), each with 100 expert demonstration episodes.

Dataset Summary

Item	Value
Robot	Franka Panda
Total Tasks	15
Episodes per Task	100
Total Episodes	1,500
FPS	30
Camera Views	3 (front, side, wrist)
Video Resolution	640×480
Video Codec	H.264 (libx264)
Data Format	LeRobot v2.1

Tasks

Manipulation Tasks (M1–M8)

ID	Task Prompt
M1	Pick up the cube and place it into the basket.
M2	Pick up the ball and place it into the basket.
M3	Place the bottle upright on the tray.
M4	Push the cube into the marked area.
M5	Pour the coffee beans from the cup into the bowl.
M6	Open the drawer, then close it.
M7	Stack the two cubes.
M8	Open the lid, then close it.

Instruction-Following Tasks (I1–I7)

ID	Task Prompt
I1	Pick up the blue cube and place it into the basket.
I2	Pick up the triangular block and place it into the basket.
I3	Pick up the largest block and place it onto the tray.
I4	Pick up the object to the left of the cylinder and place it into the basket.
I5	Pick up exactly two blocks and place them into the basket.
I6	Pick up the red cube and place it into the right basket.
I7	Place all non-red objects into the basket.

Dataset Structure

franka/
├── README.md
├── {task_name}/
│   ├── meta/
│   │   ├── info.json          # Dataset metadata
│   │   ├── episodes.jsonl     # Episode metadata
│   │   └── tasks.jsonl        # Task descriptions
│   ├── data/
│   │   └── chunk-000/
│   │       ├── episode_000000.parquet
│   │       ├── episode_000001.parquet
│   │       └── ...
│   └── videos/
│       └── chunk-000/
│           ├── image_front/
│           │   ├── episode_000000.mp4
│           │   └── ...
│           ├── image_side/
│           │   ├── episode_000000.mp4
│           │   └── ...
│           └── image_wrist/
│               ├── episode_000000.mp4
│               └── ...
└── ...

Features

Each episode frame contains:

Feature	Type	Shape	Description
action	float32	(14,)	Target joint positions + end-effector pose
observation.state	float32	(14,)	Current joint positions + end-effector pose
observation.images.image_front	video	(480, 640, 3)	Front camera view
observation.images.image_side	video	(480, 640, 3)	Side camera view
observation.images.image_wrist	video	(480, 640, 3)	Wrist camera view
timestamp	float32	(1,)	Frame timestamp
frame_index	int64	(1,)	Frame index within episode
episode_index	int64	(1,)	Episode index
index	int64	(1,)	Global frame index
task_index	int64	(1,)	Task index

Action / State Space (14-dim)

Dim	Name
0–6	Joint positions (7 DoF arm)
7	Gripper position
8–10	End-effector position (x, y, z)
11–13	End-effector orientation (rx, ry, rz)

Usage

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset(
    repo_id="AtomBench/FrankaPanda",
    root="path/to/local/frankapanda",
)

Citation

If you use this dataset in your research, please cite:

@misc{atombench,
  author = {AtomBench Team},
  title = {AtomBench FrankaPanda: A Robot Manipulation Dataset},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/AtomBench}}
}

This dataset is released under the Apache 2.0 License.