apx103 / dagger

Training a car to drive in the CarRacing-v0 Gym Environment using imitation learning.

Imitation Learning (DAgger Algorithm)

This repository contains the code for an imitation learning model and the DAgger algorithm for the CarRacing-v0 Gym Environment. This was part of a homework assignment for the Deep Reinforcement Learning course at NYU Courant taught by Lerrel Pinto.

DAgger helps the imitation learning agent learn correct actions when following sub-optimal trajectories:

Check out this paper to learn more about DAgger.

Installation

Clone this repository:

git clone https://github.com/kvgarimella/dagger.git
cd dagger

Install the requirements:

pip install -r requirements.txt

Run DAgger and train your model:

python dagger.py