main page | |
Session Title: An Orwellian approach
to AI architecture Time/Date: Friday (March 11, 2005) 9:00am - 10:00am Track: Programming Format: Lecture Experience Level: Intermediate |
|
Description: The presentation
introduces Brazil, a new approach to AI architecture. It
is similar to Hierarchical State Machine (HSM) but offers
finer granularity and higher level of decomposition
resulting into greater flexibility. Brazil is derived
from an Orwellian bureaucratic dictatorship
metaphor. Examples of bureaucratic dictatorship
are army, totalitarian state, some corporate
organizations. The name Brazil itself is derived from the
distopia movie by Terri Gilliam showing how free will is
harmful to an Orwellian society. Hence all entities in
the Brazil architecture (Agents, Dispatchers and
Collectives) lack free will completely and are
implemented as Finite State Machines. They possess
certain (usually scripted) skills allowing them to
accomplish some autonomous tasks. The entities are
organized into hierarchies and communicate via commands
and reports. Subordinate entities send reports to their
supers and receive commands from them. There is no
communication between same-level entities: whenever they
need to act cooperatively, they need to be organized
under some Dispatcher or Collective. Collectives are
Dispatchers that are also responsible for creating
subordinate entities. Dispatchers and Collectives can
form hierarchies of arbitrary depth. Each entity can
report to any number of Dispatchers but can belong to
only one Collective. Game AI itself is the top level
Collective. At the bottom of the hierarchy are Agents
with nobody reporting to them. Scripting system for
Brazil is based on simple mapping of states and event
names to the scripts in a way similar to VMT. Polymorphic
behavior of entities results from the ability to override
handlers via inheritance or dynamically. Brazil is best
applicable to the games where designers need
predictability and precise control over behaviors. Brazil
allows for building library of reusable components and
modular design of AI. While deterministic approach
prohibits emergent behavior, Brazil framework still can
emulate it to some extent. Takeaway: The audience will learn a new AI architecture applicable to wide range of games. The general methodology of Brazil can be used for designing game sub-systems besides AI as well. The underlying Orwellian metaphor provides consistent guidelines for extending and applying Brazil to many practical tasks. Intended Audience and Prerequisites: The target audience is intermediate to advanced AI programmers. Knowledge of basics of Object Oriented design, Finite State Machines and some familiarity with Hierarchical State Machine is required. |
|
main page |