Language Table

Language-Table is a suite of human-collected datasets and a multi-task continuous control benchmark for open vocabulary visuolinguomotor learning.

Installation

Installation with pip. requirements.txt contains dependencies for running the environment and simple dataset examples.

python3 -m venv ./ltvenv
source ./ltvenv/bin/activate
pip install -r ./requirements.txt
export PYTHONPATH=${PWD}:$PYTHONPATH

For running the full train script, install using requirements_static.txt, as this contains pinned versions for running the full train script.

python3 -m venv ./ltvenvtrain
source ./ltvenvtrain/bin/activate
pip install --no-deps -r ./requirements_static.txt
export PYTHONPATH=${PWD}:$PYTHONPATH

Quickstart

Examples

Scripts

Run and edit the following examples:

Load the environment and run 5 random steps:

python3 language_table/examples/environment_example.py

Load dataset and print first 5 elements:

python3 language_table/examples/dataset_example.py

Train

source ./ltvenvtrain/bin/activate
mkdir -p /tmp/language_table_train/
PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python  python language_table/train/main.py --config=./language_table/train/configs/language_table_sim_local.py --workdir=/tmp/language_table_train/

Colab

See the colab for a more complete tutorial.

Data

import tensorflow_datasets as tfds
data_directory = 'gs://gresearch/robotics/language_table/0.0.1/'
dataset = tfds.builder_from_directory(data_directory).as_dataset()

Environment

from language_table.environments import blocks
from language_table.environments import language_table
from language_table.environments.rewards import block2block

env = language_table.LanguageTable(
  block_mode=blocks.LanguageTableBlockVariants.BLOCK_8,
  reward_factory=block2block.BlockToBlockReward,
  control_frequency=10.0,
)
obs = env.reset()

Datasets

Descriptions

• Real Robot
  • language_table: 442,226 episodes of real robot relabeled data.
• Simulation (human)
  • language_table_sim: 181,020 episodes of simulation relabeled data.
  • language_table_blocktoblock_sim: 8,000 episodes of single task "block to block" data.
  • language_table_blocktoblock_4block_sim: 8,298 episodes of single task "block to block" data in the 4 block configuration.
• Simulation (oracle)
  • language_table_blocktoblock_oracle_sim: 200,000 episodes of single task "block to block" data from an oracle scripted agent.
  • language_table_blocktoblockrelative_oracle_sim: 200,000 episodes of single task "block-to-block-relative" data from an oracle scripted agent.
  • language_table_blocktoabsolute_oracle_sim: 200,000 episodes of single task "block to absolute location" data from an oracle scripted agent.
  • language_table_blocktorelative_oracle_sim: 200,000 episodes of single task "block to relative location" data from an oracle scripted agent.
  • language_table_separate_oracle_sim: 200,000 episodes of single task "separate blocks" data from an oracle scripted agent.

Summary Table

Paths

Checkpoints

Interactive Language: Talking to Robots in Real Time

Project Website  •  PDF

Corey Lynch, Ayzaan Wahid, Jonathan Tompson, Tianli Ding, James Betker, Robert Baruch, Travis Armstrong, Pete Florence

Abstract. We present a framework for building interactive, real-time, natural language-instructable robots in the real world, and we open source related assets (dataset, environment, benchmark, and policies). Trained with behavioral cloning on a dataset of hundreds of thousands of language-annotated trajectories, a produced policy can proficiently execute an order of magnitude more commands than previous works: specifically we estimate a 93.5% success rate on a set of 87,000 unique natural language strings specifying raw end-to-end visuolinguo-motor skills in the real world. We find that the same policy is capable of being guided by a human via real-time language to address a wide range of precise long-horizon rearrangement goals, e.g. "make a smiley face out of blocks". The dataset we release comprises nearly 600,000 language-labeled trajectories, an order of magnitude larger than prior available datasets. We hope the demonstrated results and associated assets enable further advancement of helpful, capable, natural-language-interactable robots.

Note

This is not an officially supported Google product.