Computing Degree Show 2016

A Dynamic Agent Framework for Tactical Position Evaluation within a Game Engine

Abstract

This project first identifies an issue in many common game agent pathfinding algorithms; paths are often pre-computed and therefore an agent cannot take into account dynamic cover, E.G a moving car or rubble of a destroyed wall, which may present a route out of the enemy sightline, and therefore a more tactically sound path. This leads to agents often exposing themselves to unnecessary danger when safer alternatives are often available. This project outlines a generic configurable agent framework for Unity3d to enable real time evaluation of enemy line of sight and positioning, giving the agent the necessary tools to move an identified goal whilst utilising static and non-static cover to best protect itself. A testing GUI framework was developed to allow for configuration of agent behaviour, map settings and enemy settings to allow for diverse testing of the agent within different scenarios. This can create new AI personalities which can be quantifiably demonstrated through the graphs generated by the GUI. The framework shows a large increase in the completion rate of agents who use this framework, as shown through quantitative evaluation of the system.
PositionEvaluationpicture