Awesome

Benchmarking Knowledge Transfer for Lifelong Robot Learning

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

[Website] [Paper] [Docs]

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. 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
• Datasets
• Getting Started
  • Task
  • Training
  • Evaluation
• Citation
• 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 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 benchmark_scripts/download_libero_datasets.py --datasets DATASET

where DATASET is chosen from [libero_spatial, libero_object, libero_100, libero_goal.

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.

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:

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.

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:

@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