Nowadays, autonomous vehicles become reality using deep learning methods, however such technology is subject to huge security challenges. Indeed, ALC (Automated Lane Centering) systems are more and more reliable, but what if the system fails and misleads the vehicle once. Hence the goal in our project was to look how ALC systems could be fooled in order to make the vehicle do an accident. This repository provides the code we used with Carla, OpenPilot, and DeepBillboard in order to achieve our semester project.
We adapted DeepBillboard to make our experiments, please take a look at their GitHub.
You will find the Github repository of Unreal Engine source code. However, this repository was made private by Epic Games so you first need to login or register freely to Epic Games and connect your GitHub account in the Connected Account Dashboard of Epic Games. Then, go on the GitHub of Epic Games and accept the invitation from Epic Games. You should now see the Unreal Engine repository.
Now that you have access to the GitHub repository, you must choose the 4.24
branch and clone it.
Then, go into the cloned folder and build it by launching the following command line.
./Setup.sh && ./GenerateProjectFiles.sh && make
You first need to clone the Carla GitHub. Once this is done, run the following command line to to get the latest assets that will be .
./Update.sh
Then, we will have to set the Unreal Engine environment variable. To do it persistently across sessions, open one of those two files.
gedit ~/.bashrc
# or
gedit ~/.profile
Then, add the following line to the bottom of the file.
make launch
Then run the following command line to start Carla with Unreal Engine.
export UE4_ROOT=~/UnrealEngine_4.24
Save the file and restart the terminal.
We first need to compile the Python API clinet that will grant control over the simulation.
make PythonAPI
Then to open Carla on Unreal Engine editor, run the following command line.
make launch
To start the server simulation, press the Play
button in Unreal Engine editor.