Using Machine Learning and Emergence to Provide Non-deterministic NPCs in Video Game Combat

Rayad Lackhan, Girendra Persaud, Alicia Layne, Eldon Marks, Department of Computer Science, University of Guyana Turkeyen Campus, Greater Georgetown, Guyana

Video games have come a long way from being pixels shifted across the screen to an ever-growing empire of graphical and storytelling masterpieces. Gameplay has, however, been regarded as the most vital aspect of a video game’s success. Shallow gameplay often revolves around poorly designed non-player characters (NPCs), which are often too predictable and break away from the intended immersive experience. Artificial intelligence is no stranger to video games, but its machine learning branch is a technology rarely explored in games due to its complexity. This research focused on providing an approach to creating non-deterministic non-player characters, within the context of video game combat, using machine learning and a combination of other scientific concepts such as emergence and biomimicry. Fightware, the implemented prototype, was designed by modelling fundamental data through biomimicry which was enabled by emergence and evolved through machine learning. The agents or ‘bots’ created were randomly assigned three of ten materials to function as limbs with their own distinct properties. These bots were then placed in combat with each other using the Naïve Bayes classifier algorithm since the model allowed for the use of any general machine learning algorithm. The knowledge base of the bots after every fight was then recorded along with a key log of the buttons they pressed. This data was then graphed and analyzed using correlation tests. It was found that 66.7% of the results were not statistically significant, which meant that there was no strong relationship between the bots with regards to their behaviour. This was further supported by the patterns displayed on line graphs which showed a minimum amount of intersections, thus proving that the model does indeed produce unique bots with non-deterministic behaviour.

Additional Abstract Information

Presenter: Rayad Lackhan

Institution: University of Guyana

Type: Oral

Subject: Computer Science

Status: Approved

Time and Location

Session: Oral 4
Date/Time: Tue 11:00am-12:00pm
Session Number: 408
List other presenters in this same room and session