This derived dataset unifies 10 LIBERO-Spatial tasks with 50 closed-loop OpenVLA rollouts per task.
video frame t is the observation before action t; the EEF/gripper state in step row t is recorded after that action.
Available: language task, 224×224 video, 7D controller command, EEF pose, gripper state, termination.
Not recorded: object poses, contact force, and per-step reward. Physical… See the full description on the dataset page:
https://huggingface.co/datasets/jiyeony/libero-spatial-openvla-rollouts-500.