Abstract:
Designing complex and realistic 3D environments for modern video games is one of
the time consuming challenges faced by current video game industry. Over the last 20
years many researches have been conducted to automate the 3D environment design.
We have critically reviewed the major approaches used in existing techniques to
automate the design of 3D environments. It was identified that current 3D
environment generation techniques being specific to one type of environment and the
lack of customizable frameworks which are common to many types of environments
as the main issues to be addressed.
According to literature, surprisingly complex and interesting global behaviours can
arise in multi agent systems as a result of simple rules that are followed by number of
simple agents operate in an environment. We hypothesize that this emergent
behaviour of multi agent systems can be used to design 3D game environments with
emergence properties that were not visible in initial constituents. This hypothesis is
inspired by the emergent behaviours of natural systems arising from simple rules that
are followed by individuals.
The proposed solution provides an approach to develop a framework which is
common to design many types of 3D game environments. Each 3D model in a 3D
game environment is associated with an agent with simple rules. Primary users of the
system would be designers of 3D video game projects and this system allows users to
introduce new 3D models and associate them with agent types. Based on the simple
rules of agents, the system arranges 3D objects in the most appropriate places in 3D
world while sensing the state of the surrounding environment. The main input for the
proposed system is a parameterized description of an imaginary 3D environment. The
design of the system consists of a multi agent system with specific agents named 3D
environment definition agent, terrain explorer agents, 3D model placing agents and
road network development agent. These agents are operating on the 3D game
environment based on simple rules and come up with a surprisingly complex
outcome. Output of the system is a 3D game environment with self organized 3D
models located and oriented in most suitable places. The final output is rendered on a
computer screen using an open source 3D game engine. The proposed approach has
been evaluated by implementing a prototype and comparing the proposed approach
with traditional 3D environment design approaches.
