README.md

miniBET

Clean implementation of conditional and unconditional behavior transformer. The API is heavily inspired by Lucidrains' implementations, and the implementation is heavily indebted to Andrej Karpathy's implementation of NanoGPT.

Installation

git clone [email protected]:notmahi/miniBET.git
cd miniBET
pip install --upgrade .

Usage

import torch
from behavior_transformer import BehaviorTransformer, GPT, GPTConfig

conditional = True
obs_dim = 50
act_dim = 8
goal_dim = 50 if conditional else 0
K = 32
T = 16
batch_size = 256

cbet = BehaviorTransformer(
    obs_dim=obs_dim,
    act_dim=act_dim,
    goal_dim=goal_dim,
    gpt_model=GPT(
        GPTConfig(
            block_size=144,
            input_dim=obs_dim,
            n_layer=6,
            n_head=8,
            n_embd=256,
        )
    ),  # The sequence model to use.
    n_clusters=K,  # Number of clusters to use for k-means discretization.
    kmeans_fit_steps=5,  # The k-means discretization is done on the actions seen in the first kmeans_fit_steps.
)

optimizer = cbet.configure_optimizers(
    weight_decay=2e-4,
    learning_rate=1e-5,
    betas=[0.9, 0.999],
)

for i in range(10):
    obs_seq = torch.randn(batch_size, T, obs_dim)
    goal_seq = torch.randn(batch_size, T, goal_dim)
    action_seq = torch.randn(batch_size, T, act_dim)
    if i <= 7:
        # Training.
        train_action, train_loss, train_loss_dict = cbet(obs_seq, goal_seq, action_seq)
    else:
        # Action inference
        eval_action, eval_loss, eval_loss_dict = cbet(obs_seq, goal_seq, None)

If you want to use your own sequence model, you can pass in that model as the gpt_model argument in the BehaviorTransformer constructor. The only extra requirement for the sequence model (beyond being a subclass of nn.Module having the input and output of the right shape) is to have a configure_optimizer method that takes in the weight_decay, learning_rate, and betas arguments and returns a torch.optim.Optimizer object.

Example task

Try out the example task on the Franka kitchen environment. You will need to install extra requirements that you can find in the examples/requirements-dev.txt file.

Fill out the details in examples/train.yaml with the paths of your downloaded dataset from here.

If you have installed all the dependencies and have downloaded the dataset, you can run the example with:

cd examples
python train.py

It should take about 50 minutes to train on a single GPU, and have a final performance of ~3.2 conditioned tasks on average.

Citation

If you use this code in your research, please cite the following papers whenever appropriate:

@inproceedings{
    shafiullah2022behavior,
    title={Behavior Transformers: Cloning $k$ modes with one stone},
    author={Nur Muhammad Mahi Shafiullah and Zichen Jeff Cui and Ariuntuya Altanzaya and Lerrel Pinto},
    booktitle={Advances in Neural Information Processing Systems},
    editor={Alice H. Oh and Alekh Agarwal and Danielle Belgrave and Kyunghyun Cho},
    year={2022},
    url={https://openreview.net/forum?id=agTr-vRQsa}
}

@article{cui2022play,
    title={From play to policy: Conditional behavior generation from uncurated robot data},
    author={Cui, Zichen Jeff and Wang, Yibin and Shafiullah, Nur Muhammad Mahi and Pinto, Lerrel},
    journal={arXiv preprint arXiv:2210.10047},
    year={2022}
}