Skip to content

Dataset · LogosRoboticsLab

Human demonstration videos and robot trajectories

LogosRoboticsLab/roboreel

Multi-view human videos and simulated robot trajectories for 10 object manipulation tasks.

What's inside

The dataset combines human demonstration video recordings and simulated robot trajectories for 10 manipulation tasks. The volume ranges from 10,000 to 100,000 items, including multi-view videos (first-person, front, left, right), depth maps, and .h5 trajectories. Episodes are split into scenarios with distractors, without distractors, and with a specified pose.

What you can do with it

  • Training imitation learning models for robotic manipulators
  • Transferring skills from human demonstration videos to robot control
  • Evaluating the robustness of robotic policies to visual distractors
Size
10K–100K rows
Downloads this month
682
Likes
2

How to load it

from datasets import load_dataset

ds = load_dataset("LogosRoboticsLab/roboreel")