Author: Jernej Habjan

Title:

• Učenje realno-časovne strateške igre z uporabo globokega spodbujevalnega učenja
• Teaching of real-time strategy game using deep reinforcement learning

• UE4 - 4.19.2
• Tensorflow-ue4 v0.8.0 for UE4.19 https://github.com/getnamo/tensorflow-ue4/releases/tag/0.8.0
• Python 3.6 64 bit
• Tensorflow 1.9
• cuda_9.0.176_win10
• cuda_9.0.176.1_windows
• cuda_9.0.176.2_windows
• cuda_9.0.176.3_windows

#Navigation

• Quick overview
  • Features
• Startup Guide
  • Visual Studio
  • Anaconda
  • Unreal Engine
  • PyTorch
  • TensorFlow GPU
• Publishing
  • Dedicated server
    • Build dedicated server and client game
    • Port forwarding and firewall
    • Check server availability
• Sources
  • Plugins used
  • Repositories used
  • Video tutorials followed
• License

Visual representation of RTS game, supported by deep reinforcement learning algorithm Alpha Zero written in python.

• Keyboard, mouse camera movement
• Minimap with optional Fog of War
• Unit creation
• Building placement
• Resource mining
• Attacking
• Unit control (control groups)
• Support for online play
• Day-night cycle

install Visual Studio

• install "Windows 8.1 sdk"
• install "Windows 10 sdk (10.0.15063.0)"

• install unreal engine from Epic Launcher or download source from GitHub
• install VaRest plugin for engine
• install Kantan Charts plugin for engine
• if source code engine has been downloaded, you have to copy plugins that are downloaded in Epic Launcher engine and paste them in source code one in Plugins folder

• Download and install python 3.6 for plugin UnrealEnginePython in

C:\Program Files\Python36

• Add it to environmental path

Install anaconda with python 3.6 -- make sure its 64 bit if using 64 bit windows

• Install it in:

C:\ProgramData\Anaconda3

• add this path under environmental variables folder

Install Python dev environment https://developer.nvidia.com/how-to-cuda-python

type this in anaconda prompt with admin rights

conda update conda
conda install numba
conda install cudatoolkit
conda install -c peterjc123 pytorch cuda90 

Install cuda:

• From programs disk install these files in this order:
• cuda_9.0.176_win10
• cuda_9.0.176.1_windows
• cuda_9.0.176.2_windows
• cuda_9.0.176.3_windows
• Extract this folder and add it to Cuda path to corresponding folders:
  • cudnn-9.0-windows10-x64-v7.1

vertify cuda installation:

• nvcc --version

• make sure its added to env variables:

- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\libnvvp

• Todo - check if i can only import whole conda environment from below tutorial
• Install tensorflow in pycharm and not from conda cmd (launch it in admin)
• Interpreter-> check "Use conda package manager button" on the right,
• add package -> tensorflow-gpu, tensorflow, tensorboard, keras 2.13 - version is important!

conda install graphviz
conda install pydot

Download Graphviz executable from Here

Add path to environmental variables and !important! restart Pycharm

C:\Program Files (x86)\Graphviz2.38\bin

to export form conda cmd

conda list --explicit > requirements.txt

to install on other machine

conda create --name myenv --file requirements.txt
activate myenv

Requirements are in _Files folder as requirements.txt

actions = ["idle", "up", "down", "right", "left", "mine_resources", "return_resources", "attack", "find_enemy", "npc", "rifle_infantry", "town_hall", "barracks", "sentry", "mining_shack", "continue_building"]

That sums up to 16 actions

• Grid of size 8x8,
• Maximum 1 unit per field.
• Units for player 1 are defined with positive numbers (1, 2, 3, 4, 5, 6...) where increased number doesn't mean nothing, but every unit is encoded with one of these numbers,
• Units for player -1 are defined with negative numbers (-1, -2, -3, -4, -5, -6...) where increased number doesn't mean nothing, but every unit is encoded with one of these numbers,
• Neutral units (minerals) are encoded with small positive integer (0.0001)
• Empty fields are represented with number 0

Point is for algorithm to learn to select town hall and execute action 'npc' before opponent does Actions:

actions = ["idle", "up", "down", "right", "left", "npc"]

Action sequence:

npc
npc
npc

Neural network must select town hall, execute action 'npc' to create workers and individually move them towards minerals and pick minerals up Actions:

actions = ["idle", "up", "down", "right", "left", "mine_resources"]

Action sequence:

npc
(while not at minerals)
    move (in apropriate direction)
gather_resources

Player would have to successfully return picked up minerals to resource drain like Town Hall or Mineral Mine. Actions:

actions = ["idle", "up", "down", "right", "left", "mine_resources", "return_resources"]

Action sequence:

npc
(while not at minerals)
    move (in apropriate direction)
gather_resources
(while not at resource drain)
    move (in apropriate direction)
return_resources

Player receives 1 TownHall and 1 NPC at start with 0 money and has to gather enough resources to build barracks. Actions:

actions = ["idle", "up", "down", "right", "left", "mine_resources", "return_resources", "barracks", "continue_building"]

Action sequence:

(while not enough minerals for barracks)
    (while not at minerals)
        move (in apropriate direction)
    gather_resources
    (while not at resource drain)
        move (in apropriate direction)
    return_resources
barracks
(while barracks not built)
    continue_building

Point is to set timeout to lower number, which would force player to build economy (building new workers) while gathering minerals. It would need to place mining_shack closer to minerals to more efficiently gather minerals. Actions:

actions = ["idle", "up", "down", "right", "left", "mine_resources", "return_resources", "npc", "mining_shack"]

Action sequence:

(while enough minerals)
    npc OR mining_shack
(while not at minerals)
    move (in apropriate direction)
gather_resources
(while not at resource drain)
    move (in apropriate direction)
return_resources

Player receives 1 TownHall and 1 NPC at start with 0 money and has to gather enough resources to build Barracks and from it Rifle unit. Actions:

actions = ["idle", "up", "down", "right", "left", "mine_resources", "return_resources", "rifle_infantry", "barracks", "continue_building"]

Action sequence:

(while not enough minerals for barracks)
    (while not at minerals)
        move (in apropriate direction)
    gather_resources
    (while not at resource drain)
        move (in apropriate direction)
    return_resources
barracks
(while barracks not built)
    continue_building
(repeat step for gathering minerals until enough for Rifle unit)
rifle_unit

Player receives 1 TownHall and 1 NPC at start with 0 money and has to gather enough resources to build Barracks and from it Rifle unit. Then has to move rifle unit towards enemy and with it keep attacking enemy units until they are destroyed. Actions:

actions = ["idle", "up", "down", "right", "left", "mine_resources", "return_resources", "attack", "find_enemy", "npc", "rifle_infantry", "barracks", "continue_building"]

Action sequence:

(while not enough minerals for barracks)
    (while not at minerals)
        move (in apropriate direction)
    gather_resources
    (while not at resource drain)
        move (in apropriate direction)
    return_resources
barracks
(while barracks not built)
    continue_building
(repeat step for gathering minerals until enough for Rifle unit)
rifle_unit
(while not at enemy)
    move (in apropriate direction)
(while enemy is alive)
    find_enemy
    (while found enemy is alive)
        attack

• Steam API now works only when publishing using "debug" or "development" option and not "shipping"!

• Make sure you don't forget to add hardware cursors to packaged game.

• If source engine is not on disk
  • download source
  • run Setup.bat
  • run GenerateProjectFiles.bat
• Switch project version to source engine
• Generate solution
• Open UE4 solution
  • Development editor - win64 build to see if there are any errors
  • close solution

• Open project normally in editor using source engine code as version
  • package project normally in win64

• Regenerate visual studio project
• Open solution
  • build using development server option to build dedicated server
  • close solution
• Copy files "Server" from TD2020/Binaries/WIn64/ (all 4) to Publish/TD2020/Binaries/Win64/
• Create shortcut of server and add at the end -log with prefix space

Fix port forwarding on your router

• Ue4

  • ip - ip of computer (local)
  • 27015 - internal port
  • 27015 - external port
  • protocol both

• Steam

  • ip - ip of computer (local)
  • 7777 - internal port
  • 7777 - external port
  • protocol both

Windows firewall - navigate to advanced settings-inbound rules - new rule - port - make 4 combinations - udp and tcp for both ue4 and steam

• Steam - View -Servers- Lan and internet
• If server is visible on lan but not on internet - check port forwarding - TODO - HAVEN'T SOLVED THIS YET
  • https://youtu.be/iMNjp5L7_oA - check comments here for solving issue with displaying on steam internet
• Open game using:

open 127.0.0.1:7777

It's located in

UE4EngineSourceCode/engine/binaries/win64/UnrealFrontend.exe

• if you don't have it - visual studio - class view - unreal frontend - right click and build, and then you can access it

• open UnrealFrontend.exe

• add custom Launch profile:

  • Project - browse to your project and select Uproject File

• Do you wish to build -try without , but if errors - yes - in development

• Cook - by the book

  • windows server
  • select en at the bottom

• Cooked maps show all - check all

• package - don't

• deploy - don't

• rename it on top:

• close this and click on newly created "launch profile"

• if errors when launching, Run it as build, or not...

Add python36.dll (64 - bit) to

YourProject/Publish/WindowsNoEditor/TD2020/Binaries/Win64/

cd C:\Users\Jernej\venv\Scripts tensorboard --logdir=C:\TrumpDefense2020\TD2020\Content\Scripts\td2020\models\logsTD2020NNet

• Kantan charts
• VaRest
• Advanced Sessions plugin - Downloaded for 4.18

• UnrealEnginePython - using source code to package project (with binary, project cannot be packaged) - Source code used on 2018.06.07
• SimpleRTS
• alpha-zero-general
• AppliedDataSciencePartners / DeepReinforcementLearning
• jdori / DeepReinforcementLearninghttps://github.com/jdori/DeepReinforcementLearning) - fixed DeepReinforcementLearning repository from AppliedDataSciencePartners for python 3

• Unreal Engine,
• pally qle,
• 3 Prong Gaming.

MIT License

Copyright (c) 2018 Jernej Habjan

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.