Libero Spatial Dataset in h5 file format for stable_wm based world model training.
Contains 2000 episodes of expert and random task execution.
Note: Current uploaded dataset is an old one genereated for dinowm training (using libero2dinowm.py), later translated to stable_wm h5 format.
To use the dataset unzip it.
To see how random actions are obtained check libero2stablewm.py
Dataset keys:
pixels
action
proprio
