short-MetaWorld (MetaWorld with 20 step trajectories)

1. Overview Short-MetaWorld is a dataset rendered from modified environment of Meta-World [1], which contains Multi-Task10(MT10) and Meta-Learning10(ML10) in total 20 tasks with 100 successful trajectories for each task. Each trajectory is padded to 20 steps.

2. File Structure This directory contains 3 sub-directories.

└── short-MetaWorld ├── task_description.py # language instructions for each task ├── img_only # rendered visual inputs for all tasks │ ├── button-press-topdown-v2 # task name │ │ ├── 0 # trajectory id
│ │ │ ├── 0.jpg # step 0 observation (224224) │ │ │ ├── 1.jpg # step 1 observation │ │ │ └── ... │ │ ├── 1 │ │ └── ... │ ├── door-open-v2 │ └── ... ├── unprocessed # trajectory actions with visual observations (256256) │ ├── unprocessed_MT10_20 # task name │ │ ├── data.pkl # a file contains all 10 tasks │ │ ├── door-open-v2.pkl # door-open task file │ │ └── ... │ └── unprocessed_ML10_20 └── r3m-processed # trajectory actions with visual obs processed by R3M [2]

1. Contact If you have any questions, please contact liangzx@connect.hku.hk

[1] Yu, Tianhe, et al. "Meta-world: A benchmark and evaluation for multi-task and meta reinforcement learning." Conference on robot learning. PMLR, 2020. [2] Nair, Suraj, et al. "R3M: A Universal Visual Representation for Robot Manipulation." Conference on Robot Learning. PMLR, 2023.