X hits on this document

PDF document

Enhanced NPC Behaviour using Goal Oriented Action Planning - page 84 / 110





84 / 110


The results of the experiments clearly highlight that GOAP is a superior AI system than FSMs aside from the technical comparison results. Modern games require that every component be as efficient and optimised as much as possible, especially for games consoles. This reason, combined with the other drawbacks listed at the end of the discussion section may begin to explain why GOAP has not yet become a more prevalent AI choice especially over FSMs. FSMs are a tried and tested AI technique and combined with their simplicity, less processing and memory overhead and relatively quicker development time mean that there may be no desire from companies to change to GOAP.

However if the GOAP could be optimised to operate at acceptable levels (F.E.A.R. was obviously able to do this) the experimental results highlight that the benefits of GOAP make it a more attractive system than the FSMs, both from a developers and player perspective. The unpredictable and superior (as shown in the results) behaviour can provide a challenging AI opponent to a player whilst the difficulty of the AI system can easily be tweaked to suit less talented players. Further tests are needed to truly gauge the system’s value to players. From a developer’s perspective, the GOAP system integrated better with the squad system, scored higher in the tests and proved to be far more flexible, maintainable, reusable and easier to manage than FSMs. The implementation and testing phases also highlighted some of the general benefits that Orkin has emphasised about the GOAP system (Orkin, 2005) i.e. dynamic re-planning and decoupling of goals and actions.

In summation, GOAP’s benefits substantially out-weigh its short-comings, it is a far superior system to FSMs and surely it is only a matter of time before more developers begin to adopt it as a primary means for controlling NPC behaviour. This dissertation has added its results and findings to the limited pool of public knowledge relating to the domain of AI planning in games. No other known project has previously taken a GOAP system and


Document info
Document views428
Page views432
Page last viewedMon Jan 23 23:26:38 UTC 2017