Skip to content

Dataset · lerobot

PushT trajectories and videos in LeRobot format

lerobot/pusht

Compact dataset of 206 episodes with video and time series for training robotic skills.

What's inside

Inside are execution records for a single control task: 206 episodes, 25,650 frames (10 fps), and 206 video files. The dataset is converted to the LeRobot framework; the robot type is not specified in the metadata. Tabular data and time series are stored in parquet format with a train split.

What you can do with it

  • Rapid debugging and testing of imitation learning algorithms
  • Training baseline control policies on the PushT simulation task
Tasks
Robotics
Size
10K–100K rows
Downloads this month
28.4K
Likes
66

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("lerobot/pusht")