Metadata-Version: 2.4
Name: hf_libero
Version: 0.1.3
Summary: LIBERO: Benchmark and environments for robot learning
Author: Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, Peter Stone
Author-email: Jade Choghari <jade.choghari@huggingface.co>
License: MIT
Project-URL: Homepage, https://github.com/huggingface/LIBERO
Project-URL: Issues, https://github.com/huggingface/LIBERO/issues
Requires-Python: >=3
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: hf-egl-probe>=1.0.1
Requires-Dist: hydra-core<1.4,>=1.2
Requires-Dist: robomimic==0.2.0
Requires-Dist: robosuite==1.4.0
Requires-Dist: bddl==1.0.1
Requires-Dist: numpy
Requires-Dist: transformers
Requires-Dist: easydict
Requires-Dist: einops
Requires-Dist: thop
Requires-Dist: matplotlib>=3.5.3
Requires-Dist: cloudpickle>=2.0.0
Requires-Dist: opencv-python
Requires-Dist: wandb
Requires-Dist: gymnasium>=0.29.0
Requires-Dist: mujoco>=3.0.0
Requires-Dist: future>=0.18.2
Dynamic: author
Dynamic: license-file
Dynamic: requires-python

<div align="center">
<img src="https://github.com/Lifelong-Robot-Learning/LIBERO/blob/master/images/libero_logo.png" width="360">


<p align="center">
<a href="https://github.com/Lifelong-Robot-Learning/LIBERO/actions">
<img alt="Tests Passing" src="https://github.com/anuraghazra/github-readme-stats/workflows/Test/badge.svg" />
</a>
<a href="https://github.com/Lifelong-Robot-Learning/LIBERO/graphs/contributors">
<img alt="GitHub Contributors" src="https://img.shields.io/github/contributors/Lifelong-Robot-Learning/LIBERO" />
</a>
<a href="https://github.com/Lifelong-Robot-Learning/LIBERO/issues">
<img alt="Issues" src="https://img.shields.io/github/issues/Lifelong-Robot-Learning/LIBERO?color=0088ff" />

## **Benchmarking Knowledge Transfer for Lifelong Robot Learning**

Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, Peter Stone

[[Website]](https://libero-project.github.io)
[[Paper]](https://arxiv.org/pdf/2306.03310.pdf)
[[Docs]](https://lifelong-robot-learning.github.io/LIBERO/)
______________________________________________________________________
![pull_figure](https://github.com/Lifelong-Robot-Learning/LIBERO/blob/master/images//fig1.png)
</div>

**LIBERO** is designed for studying knowledge transfer in multitask and lifelong robot learning problems. Successfully resolving these problems require both declarative knowledge about objects/spatial relationships and procedural knowledge about motion/behaviors. 
This repository started as a fork of the official LIBERO benchmark
. It is now maintained by the Hugging Face team, with modifications for compatibility with LeRobot
, simplified installation, and large-scale robotics experiments.
**LIBERO** provides:
- a procedural generation pipeline that could in principle generate an infinite number of manipulation tasks.
- 130 tasks grouped into four task suites: **LIBERO-Spatial**, **LIBERO-Object**, **LIBERO-Goal**, and **LIBERO-100**. The first three task suites have controlled distribution shifts, meaning that they require the transfer of a specific type of knowledge. In contrast, **LIBERO-100** consists of 100 manipulation tasks that require the transfer of entangled knowledge. **LIBERO-100** is further splitted into **LIBERO-90** for pretraining a policy and **LIBERO-10** for testing the agent's downstream lifelong learning performance.
- five research topics.
- three visuomotor policy network architectures.
- three lifelong learning algorithms with the sequential finetuning and multitask learning baselines.
---


# Contents

- [Installation](#Installation)
- [Datasets](#Dataset)
- [Getting Started](#Getting-Started)
  - [Task](#Task)
  - [Training](#Training)
  - [Evaluation](#Evaluation)
- [Citation](#Citation)
- [License](#License)


# Installtion
Please run the following commands in the given order to install the dependency for **LIBERO**.
```
conda create -n libero python=3.8.13
conda activate libero
git clone https://github.com/Lifelong-Robot-Learning/LIBERO.git
cd LIBERO
pip install -r requirements.txt
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113
```

Then install the `libero` package:
```
pip install -e .
```

# Datasets
We provide high-quality human teleoperation demonstrations for the four task suites in **LIBERO**. To download the demonstration dataset, run:
```python
python benchmark_scripts/download_libero_datasets.py
```
By default, the dataset will be stored under the ```LIBERO``` folder and all four datasets will be downloaded. To download a specific dataset, use
```python
python benchmark_scripts/download_libero_datasets.py --datasets DATASET
```
where ```DATASET``` is chosen from `[libero_spatial, libero_object, libero_100, libero_goal`.

**NEW!!!**

Alternatively, you can download the dataset from HuggingFace by using:
```python
python benchmark_scripts/download_libero_datasets.py --use-huggingface
```

This option can also be combined with the specific dataset selection:
```python
python benchmark_scripts/download_libero_datasets.py --datasets DATASET --use-huggingface
```

The datasets hosted on HuggingFace are available at [here](https://huggingface.co/datasets/yifengzhu-hf/LIBERO-datasets).

## Assets

**IMPORTANT: Asset Loading from HuggingFace Hub**

The simulation assets (3D models, textures, scene files, etc.) are now automatically loaded from HuggingFace Hub instead of being bundled with the package. When you first run LIBERO, the assets will be automatically downloaded from the Hub repository [lerobot/libero-assets](https://huggingface.co/lerobot/libero-assets) and cached locally.

This change:
- Reduces the size of the installed package
- Ensures you always have the latest assets
- Allows for easy asset versioning and updates

The assets will be cached at `~/.cache/libero/assets/` and will only be downloaded once. If you have local assets installed from a previous version, those will be used instead.


# Getting Started

For a detailed walk-through, please either refer to the documentation or the notebook examples provided under the `notebooks` folder. In the following, we provide example scripts for retrieving a task, training and evaluation.

## Task

The following is a minimal example of retrieving a specific task from a specific task suite.
```python
from libero.libero import benchmark
from libero.libero.envs import OffScreenRenderEnv


benchmark_dict = benchmark.get_benchmark_dict()
task_suite_name = "libero_10" # can also choose libero_spatial, libero_object, etc.
task_suite = benchmark_dict[task_suite_name]()

# retrieve a specific task
task_id = 0
task = task_suite.get_task(task_id)
task_name = task.name
task_description = task.language
task_bddl_file = os.path.join(get_libero_path("bddl_files"), task.problem_folder, task.bddl_file)
print(f"[info] retrieving task {task_id} from suite {task_suite_name}, the " + \
      f"language instruction is {task_description}, and the bddl file is {task_bddl_file}")

# step over the environment
env_args = {
    "bddl_file_name": task_bddl_file,
    "camera_heights": 128,
    "camera_widths": 128
}
env = OffScreenRenderEnv(**env_args)
env.seed(0)
env.reset()
init_states = task_suite.get_task_init_states(task_id) # for benchmarking purpose, we fix the a set of initial states
init_state_id = 0
env.set_init_state(init_states[init_state_id])

dummy_action = [0.] * 7
for step in range(10):
    obs, reward, done, info = env.step(dummy_action)
env.close()
```
Currently, we only support sparse reward function (i.e., the agent receives `+1` when the task is finished). As sparse-reward RL is extremely hard to learn, currently we mainly focus on lifelong imitation learning.

## Training
To start a lifelong learning experiment, please choose:
- `BENCHMARK` from `[LIBERO_SPATIAL, LIBERO_OBJECT, LIBERO_GOAL, LIBERO_90, LIBERO_10]`
- `POLICY` from `[bc_rnn_policy, bc_transformer_policy, bc_vilt_policy]`
- `ALGO` from `[base, er, ewc, packnet, multitask]`

then run the following:

```shell
export CUDA_VISIBLE_DEVICES=GPU_ID && \
export MUJOCO_EGL_DEVICE_ID=GPU_ID && \
python libero/lifelong/main.py seed=SEED \
                               benchmark_name=BENCHMARK \
                               policy=POLICY \
                               lifelong=ALGO
```
Please see the documentation for the details of reproducing the study results.

## Evaluation

By default the policies will be evaluated on the fly during training. If you have limited computing resource of GPUs, we offer an evaluation script for you to evaluate models separately.

```shell
python libero/lifelong/evaluate.py --benchmark BENCHMARK_NAME \
                                   --task_id TASK_ID \ 
                                   --algo ALGO_NAME \
                                   --policy POLICY_NAME \
                                   --seed SEED \
                                   --ep EPOCH \
                                   --load_task LOAD_TASK \
                                   --device_id CUDA_ID
```

# Citation
If you find **LIBERO** to be useful in your own research, please consider citing our paper:

```bibtex
@article{liu2023libero,
  title={LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning},
  author={Liu, Bo and Zhu, Yifeng and Gao, Chongkai and Feng, Yihao and Liu, Qiang and Zhu, Yuke and Stone, Peter},
  journal={arXiv preprint arXiv:2306.03310},
  year={2023}
}
```

# License
| Component        | License                                                                                                                             |
|------------------|-------------------------------------------------------------------------------------------------------------------------------------|
| Codebase         | [MIT License](LICENSE)                                                                                                                      |
| Datasets         | [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/legalcode)                 |
