Artificial Intelligence in Games.
Computer games are an a field of renewed interest within AI. Rather
than the static search required for games such as chess, these games
involve a simulated environment requiring the use of most existing,
as well as novel, AI techniques. The environments are challenging as
they require intelligent behaviour under uncertainty, typically in real time.
Applications of these intelligent programs range from intelligent opponents
in computer games, much sought after in the computer gaming industry,
to intelligent support units in huge systems such as the simulated battle
field environment. At least three different types of systems are of interest:
- Real time combat games, whether team games or individual.
These are the DOOM type games, in particular newer versions that
allow user-written software, such as Unreal Tournament (TM)
- Strategy games, such as Red Alert (TM).
- Team games. In particular, simulated rescue and the simulation
soccer league of RoboCup - hopefully in order to establish a team
that can participate in the annual RoboCup 2002
competition and hold its own...
Numerous approaches can be taken to achieve good performance in such
a system. In the gaming industry, the solution is to "cheat" - that is,
the agent actually acts under certainty, or has special bonuses, in which case
a simple finite-state machine works. The challenge is to actually work
with the uncertainty - preferrably using a sound probabilistic and
decision-theoretic model. These are challenging, especially in a team setting.
In addition, as the problem of evaluating such a model is hard, approximations
are neccesary in order to achieve real-time performance.
Keywords: simulated environments, reasoning under uncertainty,
real-time planning, intelligent agents, strategy and action games,
simulated battle-field.