X hits on this document

PDF document

Enhanced NPC Behaviour using Goal Oriented Action Planning - page 16 / 110





16 / 110

commercial game code and modifying it to create new custom AI behaviour. HTNs control teams of bots in a series of ‘Domination’ games in UT2005. Domination games involve teams of bots fighting over two strategic locations on a map and only one team can control a location at a time. If all the strategic locations or ‘Domination points’ are controlled by a single team for a certain period of time then the team scores a point. The results of the project highlighted that the HTN out-performed the non-HTN team and that they are well- suited to encoding strategies to coordinate teams of bots in first-person shooter games.

Another article published by Orkin discusses design decisions taken for the GOAP system and the AI agent created for F.E.A.R. (Orkin, 2004). The planning process is once more dealt with, but in a little bit more detail. Along with the prioritisation of goals, actions can also be prioritised to resolve cases where actions have the same effects. There exists no explicit mapping between goals and actions, the planner finds the valid sequence of actions to solve a goal at run-time and there is no link between goals and actions.

When looking for a solution to a goal, the planner must examine which world state symbols are unsatisfied between the agent’s current world state and the goal world state and look for actions that have effects that bring the two states closer to one another. It is possible that there can be a combinatorial explosion when searching for a valid sequence of actions for a goal but Orkin explains how the F.E.A.R. system overcame this problem by using both symbolic and non-symbolic preconditions, hashing actions by their effects and using heuristics to guide the A* search.

The issue of symbolically representing the virtual world to the agent is also addressed in this article. One of the difficulties involved with implementing a planner based on STRIPS is that STRIPS functions using Boolean symbols. This may be inadequate for a game scenario as not everything in a game world can be represented by just true of false. Before an action can be chosen by the planner, its symbolic preconditions must first be


Document info
Document views431
Page views435
Page last viewedTue Jan 24 03:30:03 UTC 2017