Note: This readme and demo was created alongside 2 AIP students

AirSim Reinforcement Learning

File Location: -- This was done because GitLab was not allowing the large file sizes.

Part 1: https://gitlab.nautilus.optiputer.net/aip_197_accchi/purple_nl/AirSim

Note: This drive is only accessible for UCSD members

Part 2: https://drive.google.com/drive/folders/1ww7ipw26vJ_Ap1Ph-0f7iSBuSxXaEm4w?usp=sharing

Installation

Installing Unreal Engine

Download the Epic Games Launcher https://www.epicgames.com/store/en-US/download

Once downloaded, install Unreal Engine 4.25 in the Library Section

Building AirSim

Install Visual Studio 2019 Make sure to select Desktop Development with C++ and Windows 10 SDK 10.0.18362

Clone the Repository on GitLab on your local computer

Follow Documentation to set up on Windows https://microsoft.github.io/AirSim/build_windows/

Follow Documentation to set up Blocks Environment https://microsoft.github.io/AirSim/unreal_blocks/

How to Use

Car Demo

Follow Documentations to set up VS solution file to run demo: https://microsoft.github.io/AirSim/unreal_custenv/

Open LandscapeMountains.sln file and set configurations to DebugGameEditor and Win64

Press F5 to run

In Window/World Settings, set GameMode Overrride to AirSimGameMode

In Edit/Editor Preferences, search for ‘CPU’ and make sure that ‘Use Less CPU when in Background’ is unchecked

Save and Press Play

Once car simulation has loaded, you can first attempt to drive the arrow keys and see that it can drive

Also a camera appears with the car simulation

Image of Car Simulation with Camera

Examples

Reinforcement Learning

Install Anaconda and with Anaconda install stable-baselines3 with command

pip install stable-baselines3[extra]

Once installed, use Anaconda command line to go into the AirSim/PythonClient/reinforcement_learning directory and run command

python dqn_car.py

There may be missing modules that cause errors when running this command and we can download these modules using pip install

Once all modules are downloaded, run the command

python dqn_car.py

If the command works, you will begin to see the car drive on its own and reset if there are obstructions

Image of Reinforcement Learning with Car Simulation

Image of Command Prompt Returns

Resources

AirSim Documentation

https://github.com/Microsoft/AirSim/#how-to-use-it

https://microsoft.github.io/AirSim/

AirSim Setup on Windows

https://microsoft.github.io/AirSim/build_windows/

AirSim Setup for Blocks Environment

https://microsoft.github.io/AirSim/unreal_blocks/

AirSim Custom Unreal Environment

https://microsoft.github.io/AirSim/unreal_custenv/

AirSim Reinforcement Learning

https://microsoft.github.io/AirSim/reinforcement_learning/