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.