compared it with any other AI system and so the findings of this project are a first in the field of GOAP in games. There are presently no open-source AI planners available for games and as a result of this project, a standalone GOAP planner was developed which in the future could be released to the general public which would allow it to be extended and improved upon.
Though the research and implementation of this project answered the research question that was posed, there is a great deal more work that can be done in the future not just continuing the work from this project, but also in the fields of GOAP and AI planning in general.
6.1 Future work
Although this project researched how GOAP can improve the AI of an NPC by testing against FSMs, another way to test if it is superior to FSMs is to perform extensive play testing. Test subjects could play against GOAP and FSM opponents and surveyed on which they felt was better. Due to the subjective nature of this test, large numbers of test subjects would be required to obtain adequate results. Ultimately, games are created to satisfy customers so this is the next logical step in determining if GOAP really is an improvement over FSMs from a player’s perspective.
Another extension of the project would be to actually modify Unreal Tournament 2005 and place the different AI agents into the real Domination, Deathmatch, Capture the Flag and Last Man Standing games modes in UT2005 rather than the 2D simulation that was created for this dissertation. Combining the systems alongside complex 3D graphics, physics and pathfinding would get far more accurate results especially from the technical experiments as they would reflect modern game requirements more closely.