Games and Artificial Intelligence Techniques Assignment 2
Students: Michael Dann and Jayden Ivanovic
Since the main focus of this assignment was machine learning, we did not spend as much time polishing the actual game as we did for assignment 1. As such, there are still a few issues / bugs we haven't fixed:
- The tree placement genetic algorithm occasionally spawns trees in the middle of lakes.
- The enemy frog occasionally gets trapped in a corner by both snakes.
- The frog never dies
If you're a student from Games and AI Techniques, please note that Jayden and I created this project many years ago in a much older Unity version. I haven't bothered fixing all the warnings about deprecated method calls and such yet, and I can guarantee that some aspects of the code are not best practice!
Please refer to the file "Report.pdf" for our write up of the game.
To run the full game that we created for the second assignment, please open the scene "MenuA2" in Unity.
When clicking play you will be presented with a loading screen. The game takes a while to initialize as the GA is generating the positions for a number of trees and flowers.
Flies can be selected by left clicking with the mouse and dragging the box over the flies you wish to move. Additionally, if you hit the numbers 1, 2, 3, 4 and 5 you will subselect that many flies based on their proximity to the current mouse position. For example, if you have 5 flies selected and press 1, you will select the closest fly to your mouse position and the others will be unselected.
Flies get points when hovering around the trees with flowers/apples (white/red trees). Return your flies to the tree in the center in order to return these points. For every 100 points you return a new fly will be spawned.
To see our setup for training the enemy frog, please open the "Training" scene.
Since training the frogs takes quite some time, we have included some pre-trained populations for demonstration purposes. To toggle between the various sample populations, click on the "GAController" GameObject and set the "Load Epoch" parameter in the "GAFrog Controller" script to a value in the set {0, 100, 200, ... , 2000} before launching the scene. (The program saves the population after every epoch but we only included multiples of 100 to keep the project's file size down.)
To train from scratch, clear out the "Load Path" parameter on the "GAController" GameObject.
There are a LOT of script components in our project, and many of them are not particularly relevant to assignment 2. The most important scripts for this assignment are as follows -
Main game:
- Assets/Scripts/DragSelectFlies.cs
- Assets/Scripts/FlyPlayerInfo.cs
Tree placement:
- Assets/Scripts/GeneticAlgs/GAController.cs
- Assets/Scripts/GenerateFoodTrees.cs
Frog training:
- Assets/Scripts/GeneticAlgs/GAFrogController.cs
- Assets/Scripts/ManagePen.cs
- Assets/Scripts/NeuralNet.cs
- Assets/Scripts/Steering/NeuralNetSteering.cs
Based on Justin's feedback from the first assignment, we avoided using GetComponent in code segments that get called every frame. However, we did not go back and fix all the GetComponents in the first assignment's code because there were sooo many and we weren't experiencing any serious frame rate issues (yet). This was a definitely a lesson learned for the next major Unity project that either of us write!