Stable-BC: Bounding Covariate Shift with Stable Behavior Cloning

This repository provides our implementation of Stable-BC in three different simulation environments and in a real world air hockey game using a 7-DoF Franka Emika robot arm. The videos for our experiments can be found [here](youtube link)

Installation and Setup

To install Stable-BC, clone the repo using

git clone https://github.com/VT-Collab/Stable_BC.git

Install the required packages and dependencies for running Stable-BC

• PyPI(pip):

pip install -r requirements.txt

• Conda Environment

conda env create -f stable_bc.yml

Implementation

The implementation details for each experiment are provided in the respective directories:

• Intersection Environment

• Quadrotor Simulation

• Vision-Based Environment

• Air Hockey Experiments

Comparison with Offline-RL

In addition to the comparison with state-of-the-art imitation learning algorithms performed in the manuscript, we also compare our proposed approach to an Offline-RL algorithm in the intersection environment. The implementation details for the Offline-RL algorithm and a detailed comparison of performance against Stable-BC is provided in:

• Intersection Environment/CQL