hildebrandt-carl/safereinforcementlearning

Self-Driving Car

This code has been adapted from the following github repo.This is part of the work done by M. Alshiekh et. al. in the paper: "Safe Reinforcement Learning via Shielding". This repo contains the following work done by me:

• A presentation summarizing the paper
• A Jupyter Notebook describing how the self-driving car example works
• An slightly modified version of the original self-driving car code found in github repo

Installing:

This repository was tested on a fresh install of Ubuntu 18.04. First install the required dependencies:

$ sudo apt-get install python3-pip
$ sudo apt-get install cmake
$ sudo apt-get install swig
$ sudo apt install python3-notebook jupyter jupyter-core python-ipykernel
$ pip3 install jupyter
$ pip3 install -r requirements.txt
$ pip3 install matplotlib

For notes on the changes made to the original codebase please look inside the Jupyter Notebook.

Running the Code:

You can run the code using the following commands:

## No Shield
$ python3 krl.py -v

## Big Negative Reward
$ python3 krl.py -v -bn

## Preemtive Shield
$ python3 krl.py -v -p

## Post Shield
$ python3 krl.py -v -s

You can then plot the results using:

$ python3 plot_score.py

Running the Jupyter Notebook

You can see the Jupyter Notebook here.

Launch the notebook $ jupyter notebook