README.md

metadata

task_categories:
  - robotics

bridge dataset

version 1.0.0 consists of 60K trajectories in RLDS format

To use:

import tensorflow_datasets as tfds
import tqdm
import argparse

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument("--rlds_dir", type=str, default="bridge_data/1.0.0")
    args = parser.parse_args()

    ds_builder = tfds.builder_from_directory(args.rlds_dir)
    dataset = ds_builder.as_dataset(split='all')

    ds_length = len(dataset)
    dataset = dataset.take(ds_length)
    it = iter(dataset)

    for i in tqdm.tqdm(range(ds_length)):
        episode = next(it)
        print("episode: ", i)
        steps = episode['steps']
        print("key in a traj: ", episode.keys())

        for j, step in enumerate(steps):
            # print(step['observation'].keys())
            print(f" [step {j}] action: ", step["action"])
            print(f" [step {j}] state: ", step['observation']['state'])