RonEnix

Simple Co-operative Agents in a Party-based game application.

Contents

1. Brief Description and Aim
2. Details and Implementations
3. Media
4. Code
5. Report

---

BRIEF DESCRIPTION AND AIM

A project that works with Neural Networks and Genetic Algorithms to simulate learning agents. These agents are taught to learn specific behaviours to adapt to the designed simulation.

This simulation is in a form of a party-base co-operative effort, which multiple agents use the behaviours they have learnt to the most advantageous results when put forth against a hard and difficult enemy.

How this approach came to existence was a result of a well know feature in many massively-multi-player online games to date. One that has a large influence to this design was World of Warcraft. One of the system that involves around this game is in the form of a party-vs-enemy encounters where a group of 5 players undergo a “run” through various dungeons where enemies are a higher difficulty than the usual type of enemies that can be found roaming around the persistent world. These enemies are so known as “Elites”.

One of the other feature that’s used within this project is in the form of “threat”. Every enemy in the game has a gauge that measures the threat level of each character. This threat amount increases for each player when they do damage or use healing. The threat amount allows the enemy to determine who to attack there on.

This project works on these mechanics and apply learning agents in the process. Each agents will be given a particular role to work on within the group. This brings forth 3 main character game classes: Healers, Defenders (tank), and Damage dealers.

The functionality/learning on each of the agents will be controlled by a learning algorithm. This approach uses Neuro-Evolution of Augmenting Topologies, in short NEAT. This algorithm works with Neural networks and genetic algorithms, allow networks to evolve with their weight modified through the fitness levels the networks attain during a particular cycle or simulation.

---

DETAILS AND IMPLEMENTATIONS

more details on this section soon…

---

MEDIA

Screenshots

Video

coming soon…

---

CODE

The project source code and the code documentation with it can be found here:
ANN Level 3 Project (53 MB compressed – 160MB).
The zip file also includes the report documentation.

You can access the code documentation here: http://ronenix.com/contents/ANN/html/index.html
Documentation created using Doxygen and contains the actually function code implementation in the documentation.

Version 1.0.1

---

REPORT

Access to the project report can be found here: Report.doc