Wjjjh/libero_lingbot_va
LIBERO datasets pre-encoded for LingBot-VA This repository mirrors three community-preprocessed LIBERO suites for easier migration and training with the official Robbyant/LingBot-VA pipeline. Contents libero_spatial/: 500 episodes, 10 tasks, LeRobot v2.1, dual-camera Wan2.2 latents. libero_goal/: 500 episodes, 10 tasks, LeRobot v2.1, dual-camera Wan2.2 latents. libero_object/: 500 episodes, 10 tasks, LeRobot v2.1, dual-camera Wan2.2 latents. Each suite contains… See the full description on the dataset page: https://huggingface.co/datasets/Wjjjh/libero_lingbot_va.
LIBERO datasets pre-encoded for LingBot-VA
This repository mirrors three community-preprocessed LIBERO suites for easier migration and training with the official Robbyant/LingBot-VA pipeline.
Contents
libero_spatial/: 500 episodes, 10 tasks, LeRobot v2.1, dual-camera Wan2.2 latents.libero_goal/: 500 episodes, 10 tasks, LeRobot v2.1, dual-camera Wan2.2 latents.libero_object/: 500 episodes, 10 tasks, LeRobot v2.1, dual-camera Wan2.2 latents.
Each suite contains per-episode action/state Parquet files, meta/episodes.jsonl with action_config, dual-camera precomputed latent .pth files, and empty_emb.pt.
Upstream sources
- Spatial: armanakbari4/libero-spatial-latents-v2
- Goal: armanakbari4/liberogoal_preencoded
- Object: armanakbari4/liberoobject_preencoded
- Original benchmark: LIBERO
The data was validated locally with the official LingBot-VA LatentLeRobotDataset: all three suites expose 500 episodes, 500 Parquet files, and 1,000 dual-camera latent files per suite.
Please follow the licenses and attribution requirements of the upstream LIBERO dataset and preprocessing repositories.