Post on 10-May-2015
description
Preliminary Results
Depto. Arquitectura y Tecnología de Computadores
Universidad de Granada (Spain)
Contact: ANTONIO M. MORA (amorag@geneura.ugr.es)
What’s/Why Unreal?
Is a First Person Shooter (FPS) by Epic Games.
From Unreal in 1998 to Unreal Tournament 3 in 2007.
Very good AI for the standard bots (autonomous players).
Open programming environment (UnrealEd) and native
language (UnrealScript).
Instituto Tecnológico de Informática
Universidad Politécnica de Valencia (Spain)
Unreal (1998) Unreal Tournament 3 (2007)
ROAMING State (Substates and flow lines)
UnrealEd (UnrealScript)
AI in Unreal
Each bot follows a finite state machine (FSM) with plenty
of states and substates.
The transition in the FSM depends on some parameters
(usually hard-coded) which models the status, location in the
map (scenario) or relationship with the enemies.
The whole FSM (or just a part/state) can be modeled as a
tree of rules.
Genetic Bots (G-Bots)
Genetic Algorithm
based bot (GA-Bot),
which evolves a set of
parameters.
Genetic Programming
based bot (GP-Bot),
which evolves the set of
rules which defines the
state transitions.
DEPENDS ON
MODELS
GA
EVOLUTIONARY
PROCESS
PG
EVOLUTIONARY
PROCESS
Standar
d AI
Standar
d AI
Standar
d AI
popula
tion
popula
tion
Evalu
ati
on
(F
itn
ess
calc
ula
tio
n)
Evaluation Function
Each individual in the population is an AI ‘representation’ (a set
of parameters or rules), which models a behaviour.
The fitness calculation is performed by assigning the individual
as the G-Bot’s AI, and placing it into a scenario to fight against
some standard (AI) bots.
The fitness will be a combination of the number of killed
enemies, times the bot has been defeated, collected items and
weapons, and life time.