Large LeRobot dataset with 87212 Google robot episodes across 599 tasks.
A robotics dataset generated with LeRobot v2.0, containing 87212 episodes and 3786400 frames from google_robot across 599 tasks at 3 FPS.
Useful for training and evaluating imitation learning or robot policy models on manipulation tasks.
Use the 87k robot episodes to train visuomotor policies that map camera observations to actions for the 599 covered tasks.
Evaluate generalization across 599 tasks using the single train split of 3.7M frames recorded from a Google robot.
Leverage the 87k provided videos to build models that predict task sequences or detect task completion from low-FPS footage.
from datasets import load_dataset
ds = load_dataset("IPEC-COMMUNITY/fractal20220817_data_lerobot")A large robotics dataset of 87,212 episodes (3.7M frames) from a Google robot covering 599 tasks, stored in Parquet with videos.
Verified reviews from the community shape this listing's rating.
Loading reviews…