initial commit

.flake8

@@ -0,0 +1,13 @@
+[flake8]
+exclude = .git
+max-line-length = 88
+select = E,F,W,C
+ignore=W503,
+       E203,
+       E731,
+       E722,
+       F841,
+       E402,
+       E741,
+       E501,
+       C406,

.github/workflows/pre-commit.yaml

@@ -0,0 +1,14 @@
+name: pre-commit
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+jobs:
+  pre-commit:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v3
+    - uses: actions/setup-python@v3
+    - uses: pre-commit/action@v3.0.0

.gitignore

@@ -0,0 +1,168 @@
+tests/debug_dataset/bridge_dataset/1.0.0/dataset_statistics_*.json
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+.ipynb_checkpoints/
+wandb
+*.png
+*.sif
+.vscode
+.idea
+datasets/debug_dataset/bridge_dataset/1.0.0/action_*

.pre-commit-config.yaml

@@ -0,0 +1,40 @@
+default_language_version:
+    python: python3.10
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v2.3.0
+    hooks:
+      - id: check-yaml
+      - id: check-ast
+      - id: check-added-large-files
+        exclude: ^examples/
+      - id: check-case-conflict
+      - id: check-merge-conflict
+      - id: end-of-file-fixer
+      - id: trailing-whitespace
+      - id: detect-private-key
+      - id: debug-statements
+        exclude: ^experiments/
+  - repo: https://github.com/psf/black
+    rev: 22.10.0
+    hooks:
+      - id: black
+        exclude: ^experiments/
+  - repo: https://github.com/PyCQA/flake8
+    rev: 6.1.0
+    hooks:
+      - id: flake8
+        exclude: ^experiments/
+  - repo: https://github.com/pycqa/isort
+    rev: 5.12.0
+    hooks:
+      - id: isort
+        exclude: ^experiments/
+        args: ["--profile", "black", "--src", "crossformer", "--src", "experiments"]
+  - repo: https://github.com/srstevenson/nb-clean
+    rev: 3.1.0
+    hooks:
+      - id: nb-clean
+        args:
+          - --remove-empty-cells
+          - --preserve-cell-outputs

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2023 Robotic AI & Learning Lab Berkeley
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,88 @@
+# Scaling Cross-Embodied Learning: One Policy for Manipulation, Navigation, Locomotion and Aviation
+<!-- [![arXiv](https://img.shields.io/badge/arXiv-XXXX-df2a2a.svg)]() -->
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/rail-berkeley/crossformer/inference_pretrained.ipynb)
+[![HF Models](https://img.shields.io/badge/%F0%9F%A4%97-Models-yellow)](https://huggingface.co/rail-berkeley/crossformer)
+[![Python](https://img.shields.io/badge/python-3.10-blue)](https://www.python.org)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+[![Static Badge](https://img.shields.io/badge/Project-Page-a)](https://crossformer-model.github.io/)
+
+[Ria Doshi](https://www.linkedin.com/in/riadoshi/), [Homer Walke](https://homerwalke.com/), [Oier Mees](https://www.oiermees.com/), [Sudeep Dasari](https://sudeepdasari.github.io/), [Sergey Levine](https://people.eecs.berkeley.edu/~svlevine/)
+<hr style="border: 2px solid gray;"></hr>
+
+This repo contains code for training and finetuning CrossFormer. CrossFormer is a transformer-based robot policy trained on 900K robot trajectories across 20 different robot embodiments. Our codebase is built on the [Octo codebase](https://github.com/octo-models/octo).
+
+![CrossFormer model](docs/assets/teaser.jpg)
+
+## Get Started
+
+Follow the installation instructions, then load the pre-trained CrossFormer model! See [our colab notebook](https://githubtocolab.com/rail-berkeley/crossformer/inference_pretrained.ipynb) for an inference example.
+
+```python
+from crossformer.model.crossformer_model import CrossFormerModel
+model = CrossFormerModel.load_pretrained("hf://rail-berkeley/crossformer")
+print(model.get_pretty_spec())
+```
+
+Out of the box, CrossFormer can control single and dual arm manipulation systems, wheeled robots, quadcopters, and quadrupeds, and can be instructed via language commands or goal images.
+CrossFormer uses a modular attention structure in its transformer backbone, allowing it to be effectively finetuned to robot setups with new sensory inputs, action spaces, and morphologies, using only a small target domain dataset and accessible compute budgets.
+
+
+## Installation
+```bash
+conda create -n crossformer python=3.10
+conda activate crossformer
+pip install -e .
+pip install -r requirements.txt
+```
+For GPU:
+```bash
+pip install --upgrade "jax[cuda11_pip]==0.4.20" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
+```
+
+For TPU
+```bash
+pip install --upgrade "jax[tpu]==0.4.20" -f https://storage.googleapis.com/jax-releases/libtpu_releases.html
+```
+See the [Jax Github page](https://github.com/google/jax) for more details on installing Jax.
+
+## Checkpoint
+
+You can find the pre-trained CrossFormer 130M parameter checkpoint [here](https://huggingface.co/rail-berkeley/crossformer).
+
+## CrossFormer Pre-training
+
+To reproduce CrossFormer pre-training, edit [scripts/configs/pretrain_config.py](scripts/configs/pretrain_config.py) to point to your data and log directory. Then, run:
+
+```bash
+python scripts/train.py --config scripts/configs/pretrain_config.py
+```
+
+To download the pre-training datasets from the [Open X-Embodiment Dataset](https://robotics-transformer-x.github.io/),
+install the [rlds_dataset_mod package](https://github.com/kpertsch/rlds_dataset_mod)
+and run the [prepare_open_x.sh script](https://github.com/kpertsch/rlds_dataset_mod/blob/main/prepare_open_x.sh).
+
+Pre-training takes 47 hours on a TPUv5-256 pod.
+
+## CrossFormer Finetuning
+
+To run finetuning on your own dataset, convert your dataset to the RLDS format using [this repository](https://github.com/kpertsch/rlds_dataset_builder). Then, edit [scripts/configs/finetune_config.py](scripts/configs/finetune_config.py), and run:
+
+```bash
+python scripts/finetune.py --config scripts/configs/finetune_config.py
+```
+
+There are a few options for finetuning CrossFormer. If your dataset has an observation space and action space that was used during pre-training, you can finetune from entirely pre-trained weights, using the existing observation tokenizers, action heads, and transformer backbone. Otherwise, you can initialize new observation tokenizers and/or action heads while keeping the pre-trained transformer backbone. Additionally, you may choose to finetune the entire model or freeze the transformer and finetune only the action head. Finally, you can choose to finetune on your data with goal image conditioning, language conditioning, or both.
+
+See the comments in [scripts/configs/finetune_config.py](scripts/configs/finetune_config.py) for an explanation of how to configure these fine-tuning options.
+
+
+## Citation
+
+```bibtex
+@article{doshi24-crossformer,
+    title={Scaling Cross-Embodied Learning: One Policy for Manipulation, Navigation, Locomotion and Aviation},
+    author={Ria Doshi and Homer Walke and Oier Mees and Sudeep Dasari and Sergey Levine},
+    journal={arXiv preprint arXiv:id},
+    year={2024}
+}
+```