metadata
dataset_name: LIBERO-Mem
pretty_name: 'LIBERO-Mem: Long-Horizon Object-Centric Kitchen Manipulation'
tags:
- robotics
- imitation-learning
- reinforcement-learning
- manipulation
- vision
- hdf5
- datasets
license: mit
task_categories:
- reinforcement-learning
- other
language:
- en
LIBERO-Mem-Raw Dataset Specification
.hdf5 demonstration files — simulator initial states, proprioception, and control actions
Please refer to this LIBERO-Mem-Raw repo for downloading the raw setting to obtain initial states (120 trajs per task) for evaluation setup.
Otherwise, please refer to LIBERO-Mem for the full training setting (100 trajs per task).