/chicken

Did the chicken cross the road? Answered through Inverse Reinforcement Learning!

Primary LanguageJupyter Notebook

The passive chicken and aggressive car problem

Objective: Inducing passive-aggressive behavior in self-driving cars using Inverse Reinforcement Learning (IRL), specifically Behavioral Cloning (BC). We use Gym-Duckietown, a self-driving car simulator for our experiments.

Framework, langauge, OS: Pytorch >= 0.4.1, Python 3.7, Windows 8.1

Prerequisites

The Gym-Duckietown repo has been cloned into the "PedestrianSimulation" folder. Please follow the instructions on the repo to install all requirements

Training

Add "PedestrianSimulation" to path in one of two ways:

#From command line
export PYTHONPATH="${PYTHONPATH}:`pwd`"

#From ipython/python
import sys
sys.path.append('chicken/PedestrianSimulation')

Reinforcement learning: code at PedestrianSimulation/learning/reinforcement/pytorch
python -m reinforcement.pytorch.train_reinforcement

Imitation learning: code at PedestrianSimulation/learning/imitation/pytorch
python -m imitation.pytorch.train_imitation

Rendering Environment

To see the simulated pedestrain in the "PedestrianSimulation" folder
python manual_control.py

README last updated with instructions: Aug 2019