WeiChow/VLATrainingDataset

Open X-Embodiment Dataset (unofficial)

RLDS dataset for train vla

use this dataset

download the dataset by hf: (<THE DATASET NAME> can be bridge_orig, dorid,fmb,...etc.)

cd <PATH TO BASE DATASETS DIR>
huggingface-cli download WeiChow/VLATrainingDataset --include <THE DATASET NAME>/* --local-dir . --local-dir-use-symlinks False

prepare by yourself

The code modified from rlds_dataset_mod

We upload the precessed dataset in this repository ❤

below is the code for processing ⚙

prepare gsutil

# https://cloud.google.com/sdk/docs/install-sdk?hl=zh-cn#linux
curl -O https://dl.google.com/dl/cloudsdk/channels/rapid/downloads/google-cloud-cli-linux-x86_64.tar.gz
tar -xf google-cloud-cli-linux-x86_64.tar.gz
./google-cloud-sdk/install.sh
./google-cloud-sdk/bin/gcloud init --console-only
./google-cloud-sdk/bin/gcloud components install gsutil
# check gsutil
export PATH=$PATH:/path/to/google-cloud-sdk/bin
ls ./google-cloud-sdk/bin/gsutil

prepare environment

conda env create -f environment_ubuntu.yml
conda activate rlds
mkdir data_download
mkdir data_tmp

All the rights are reserved.