Episodes Preview anvil_openarm Visualizer
40 episodes · 60 fps · 4 cameras · 640×480 h264

0605-t-foam

This dataset was created using LeRobot.

Dataset Description

A bimanual manipulation dataset collected on the Anvil OpenArm robot.

  • Robot type: anvil_openarm
  • Task: manipulation
  • Codebase version: v3.0
  • FPS: 60
  • Total episodes: 40
  • Total frames: 51,039
  • Total tasks: 1
  • Splits: train (episodes 0–40)

Camera Views

The dataset includes four synchronized camera streams, each 640×480 H.264 video at 60 fps:

Key Description
observation.images.chest Chest-mounted camera
observation.images.waist Waist-mounted camera
observation.images.wrist_l Left wrist camera
observation.images.wrist_r Right wrist camera

Robot State & Action Space

The robot is a dual-arm system with 16 DoF total (7 joints + 1 finger joint per arm).

  • observation.state — float32[16] joint positions
  • observation.velocity — float32[16] joint velocities
  • observation.effort — float32[16] joint efforts
  • action — float32[16] target joint positions

Joint order (both state and action):

left_finger_joint1, left_joint1..left_joint7,
right_finger_joint1, right_joint1..right_joint7

Usage

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset("DMDphilai/0605-t-foam")

Citation

@misc{lerobot,
  author = {Cadene, Remi and others},
  title = {LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch},
  year = {2024},
  url = {https://github.com/huggingface/lerobot}
}
Downloads last month
90