RW-RL-HIL-Dataset
Description
RW-RL-HIL-Dataset contains 82.23 hours, 3,347 episodes, and 4,440,155 frames of R1Lite household manipulation across nine tasks during policy deployment. Per-frame control-authority flags and intervention boundaries distinguish autonomous actions from human corrections: 3,068 episodes contain interventions, totaling 9.39 hours across 6,408 segments; 279 episodes contain none. The LeRobot v2.1 release pairs head and wrist RGB at 15 fps with state and actions. It supports failure detection and recovery learning, but its mixed control stream and intervention-biased collection require appropriate filtering.
License
CC BY 4.0; commercial use, adaptation, and redistribution are permitted with attribution.*
* Double-check the publisher's current license and usage terms before using this dataset.