ir-lab/modattn

Codebase for CoRL 2022 paper "Modularity through Attention: Efficient Training and Transfer of Language-Conditioned Policies for Robot Manipulation"

ModAttn

Codebase for CoRL 2022 paper "Modularity through Attention: Efficient Training and Transfer of Language-Conditioned Policies for Robot Manipulation"

In this paper, we propose a sample-efficient approach for training language-conditioned manipulation policies that allows for rapid training and transfer across different types of robots. By introducing a novel method, namely Hierarchical Modularity, and adopting supervised attention across multiple sub-modules, we bridge the divide between modular and end-to-end learning and enable the reuse of functional building blocks.

The full experiments include the following steps:

• Data collection from the simulator
• Training the model
• Deploying the trained model back onto the simulator as an agent

This repo mainly aims at providing the code needed for step 2: training the model. Expert demonstrations are required to run this repo. After running this repo, you should expect to have a trained model which is ready to act as an agent.

Already Supported Simulators v.s. Your Own Simulator

In this repo, we provide implementions of dataloaders and training scripts of the following 2 simulators. After you collect the dataset using the code provided in the simulators' repos, you are good to start the training using this repo.

Mujoco simulator with Panda, Kinova and UR5 robots

TinyLanguageRobots 2D robot simulator for manipulation tasks

Bring Your Own Simulator

It is also possible to plug in any of your own simulators. In such case, you just need to provide the following items to run the training:

• Your expert demonstrations
• Your own dataset class and dataloader
• Customization of the training script to import your own demos and dataloaders.

This repo mainly aims at providing the code needed for step 2: training the model. Expert demonstrations are required to run this repo. After running this repo, you should expect to have a trained model which is ready to act as an agent.

Python Environment Setup

git clone https://github.com/ir-lab/ModAttn.git
pip install -r requirements.txt

Training

There are main*.py files in the root folder, which are the scripts to start the training. Each main*.py file will import the model file and dataset file for the training. If you would like to use your own dataset, just import the dataset class to a main*.py file. The following are training scripts for the datasets of the 2 already supported simulators:

Train the Mujoco robots

python main_mujoco_robot.py /path/to/train/set /path/to/val/set /path/to/ckpt/folder

Train the TinyLangRobot robot

python main_tiny_lang_robot.py /path/to/train/set /path/to/val/set /path/to/ckpt/folder

More Visualization of Results

Sim2real Transfer

Transfer across robots

Obstacle Avoiding

Inspection of network by visualizing the output of each sub-modules

BibTeX

@inproceedings{
    zhou2022modularity,
    title={Modularity through Attention: Efficient Training and Transfer of Language-Conditioned Policies for Robot Manipulation},
    author={Yifan Zhou and Shubham Sonawani and Mariano Phielipp and Simon Stepputtis and Heni Ben Amor},
    booktitle={6th Annual Conference on Robot Learning},
    year={2022},
    url={https://openreview.net/forum?id=uCaNr6_dQB0}
}

Contact

Feel free to contact us if you have any questions!

Yifan Zhou

modattn's Issues

Any plans for releasing the evaluation code?

Hi,

Thank you for your interesting work and releasing the training code. I was wondering if there are any plans to release the evaluation code?

Since the evaluation part is quite specific (as the action space consists of DMPs and not just low level end-effector/joint actions). It would be extremely helpful if the evaluation code can be provided.

Thanks,
Mohit Sharma