This repository contains the implementation of gaze-regularized imitation learning. The trained model is tested in AirSim with a quadrotor to perform tasks like search and navigation autonomously. At present, the repository contains implementation of GRIL with baseline algorithms. The GIF below shows GRIL's generalization on a moving target scenario.
The code has been tested on Ubuntu 18.04. The development environment is based on Python3 and Tensorflow 2.0+. This environment can be setup in a virtual environment using the requirements file included in the repository. After creating virtual environment, run the following.
pip3 install -r requirements.txtPaper: Imitation Learning with Human Eye Gaze via Multi-Objective Prediction ILHF Workshop ICML 2023 Video: Quadrotor following a moving truck