Skip to content

BerkeleyAutomation/IIFL

Repository files navigation

IIFL: Implicit Interactive Fleet Learning

Code for the following paper:

G. Datta*, R. Hoque*, A. Gu, E. Solowjow, K. Goldberg. IIFL: Implicit Interactive Fleet Learning from Heterogeneous Human Supervisors. Conference on Robot Learning (CoRL), 2023.

Installation

Installation instructions are similar to the IFL Benchmark on Github. First create a Python 3.8 virtual environment and install dependencies by running . install.sh.

To run the IFL Benchmark you will need to install Isaac Gym. Download Isaac Gym 1.0rc3 from https://developer.nvidia.com/isaac-gym (you may need to send a request but it should be quickly approved) and read the installation instructions in the docs to pip install into the virtual environment. You will need NVIDIA driver version >= 470.

Then clone NVIDIA IsaacGymEnvs from https://github.com/NVIDIA-Omniverse/IsaacGymEnvs and pip install it into the virtual environment. Note: make sure to run git checkout 347cfbfaeeb708e7e94bc3bd8e7f2ef069e24fde for the correct version of IsaacGymEnvs (1.3.0), since IsaacGymEnvs is actively under development.

Reproducing Results

Simply run

. scripts/run_[env].sh

where env is one of {ant, anymal, ball_balance, franka_cube}. This will run with default expert checkpoints and offline datasets, which you can re-generate if you wish.

Acknowledgement

IFL implementation is based on the IFL Benchmark. IBC implementation is adapted from Kevin Zakka's PyTorch implementation.

About

Implicit Interactive Fleet Learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published