li-whale/il_gym0

IL Gym 0 Dataset

This is a robotics dataset for imitation learning collected using the LeRobot framework.

Dataset Information

• Total Episodes: 30
• Total Frames: 2342
• Total Tasks: 1
• Total Videos: 60
• FPS: 10
• Codebase Version: v2.1

Dataset Structure

This dataset contains:

• State observations: 18-dimensional state vectors
• Actions: 4-dimensional action vectors (delta_x_ee, delta_y_ee, delta_z_ee, gripper_delta)
• Visual observations: Front and wrist camera images (128x128, 3 channels)
• Rewards and done signals
• Complementary information: discrete penalty values

Features

• observation.state: Float32 array of shape [18] containing state information
• action: Float32 array of shape [4] containing end-effector deltas and gripper commands
• observation.images.front: Video data from front camera (128x128x3, AV1 codec, 10 FPS)
• observation.images.wrist: Video data from wrist camera (128x128x3, AV1 codec, 10 FPS)
• next.reward: Float32 reward signal
• next.done: Boolean done signal
• complementary_info.discrete_penalty: Additional penalty information

Usage

This dataset can be used with the LeRobot framework for training imitation learning policies.

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset("il_gym0")

Citation

If you use this dataset, please cite the LeRobot framework.