3/28/15

AI Architectures
Gustavo Patow
IMAE - UdG
Par7ally based on
AI Architectures: A Culinary Guide
h>p://intrinsicalgorithm.com/IAonAI/tag/hfsm/

Mexican Food Analogy

The outside is a content delivery mechanism

3/28/15

Bunch of rules
Simply adding rules here or there around our code
that change the direcBon of things in a fairly
haphazard manner
Obviously, the problem with that is you can only get
so much content before things become unstable
Its a bit unwieldy as well
Most importantly, every Bme you take a bite of your
content, you never know when the enBre plaKorm is
going to simply fall apart

The most basic architecture

Adding a liLle bit of structure to a
bunch of otherwise disjointed rules
maps over somewhat to the most basic
of AI architectures

3/28/15

The nite state machine (FSM)

The most basic part of a
FSM is a state:

an AI agent is doing or
being something at a
given point in Bme
It is said to be in a state
TheoreBcally, an agent
can only be in one state
at a Bme (although other
implementaBons exist)

Unaware

Combat

Investigate

Search

The nite state machine (FSM)

This organizes the agent behavior beLer
Because everything the agent needs to
know about what it is doing is contained
in the code for the state that it is in
The animaBons it needs to play to act out a
certain state, for example, are listed in the
body of that state.

3/28/15

An example
The transiBon logic in any given state may be as simple or as
complex as needed
For example, it may simply involve a countdown Bmer that
says to switch to a new state aVer a designated amount of
Bme

It may be a simple random chance that a new state should be

entered
For example, State A might say that, every Bme we check, there
is a 10% chance of transiBoning to State B
We could even elect to make the new state that we will
transiBon to a result of a random selecBon as wellsay a 1/3
chance of State B and a 2/3 chance of State C.

Another one
More commonly, state machines employ elaborate trigger
mechanisms that involve the game logic and situaBon
For instance our guard state may have the logic, if [the player
enters the room] and [is holding the Taco of Power] and [I have the
Salsa of SmiBng], then aLack the player
at which point my state changes from guard to a>ack.
Note the three individual criteria in the statement

We could certainly have a dierent statement that says, if [the

player enters the room] and [is holding the Taco of Power] and [I
DO NOT have the Salsa of SmiBng], then ee

Obviously, the result of this is that I would transiBon from guard to

ee instead

3/28/15

Problems
First, as the number
of states increases,
the number of
potenBal transiBons
increases as well
at an alarming rate!

Workload for each new state

To have that state accessible, go and
touch every single other state that
could potenBally transiBon to it
To add a State E, we would have to
edit A-D to add the transiBons to E
EdiBng a states logic invokes the
same problem
You have to remember what other
states may be involved and revisit
each one

3/28/15

Workload
AddiBonally, any logic that
would be involved in that
transiBon must also be
interworked into the other
state-specic logic that may
already be there
Suers from some of the same
fragility of the ad hoc bunch of
rules we menBoned earlier

The Behavior Tree

It is useful to point out the dierences

between an ac7on and a decision
In the FSM, our agents were in one state at a Bme
they were doing something at any given moment
even if that something was doing nothing

Inside each state was decision logic that told them if they should change
to something else and, in fact, what they should change to
That logic oVen has very li>le to do with the state contained in and more
to do with what is going on outside the state or even the agent itself
If I hear a gunshot, it really doesnt maLer what Im doing at the BmeIm going to
inch, duck for cover, wet myself, or any number of other appropriate responses.
Therefore, why would I need to have the decision logic for React to Gunshot in each
and every other state I could have been in at the Bme?

3/28/15

The Behavior Tree

Separates the states from the decision logic

The Behavior Tree

States and decision logic sBll exist in the
AI code
but they are not arranged so that the
decision logic is in the actual state code

Instead, the decision logic is removed to a

stand-alone architecture
This allows it to run by itselfeither
conBnuously or as neededwhere it
selects what state the agent should be in
All the state code is responsible for is
doing things that are specic to that state
such as animaBng, changing values in the
world, etc.

3/28/15

The Behavior Tree

All the decision logic

is in a single place!!!

Hierarchical Finite State Machine

some states contain other related states

making the organizaBon more manageable

3/28/15

Hierarchical Finite State Machine

There are mulBple levels of states
Higher-level states will only be concerned
with transiBoning to other states on the
same level
On the other hand, lower-level states
inside the parent state can only transiBon
to each other
This Bered separaBon of responsibility
helps to provide a li>le structural
organiza7on to a at FSM and helps to
keep some of the complexity under
control

Planner
Like a behavior tree, the reasoning architecture behind a
planner is separate from the code that does stu
While the end result of a planner is a state (just like the FSM
and behavior tree), how it gets to that state is signicantly
dierent
A planner compares its situaBonthe state of the world at
the momentand compares it to a collecBon of individual
atomic acBons that it could do
It then assembles one or more of these tasks into a sequence
(the plan) so that its current goal is met

3/28/15

Planner

A planner actually works

backwards from its goal

Planner

For example, if the goal is kill player, a planner might

discover that one method of saBsfying that goal is to shoot
player.

Of course, this requires having a gun.

If the agent doesnt have a gun, it would have to pick one up.
If one is not nearby, it would have to move to one it knows exists.
If it doesnt know where one is, it may have to search for one.
The result of searching backwards is a plan that can be executed
forwards.

Of course, if another method of saBsfying the kill player

goal is to throw a Taco of Power at it, and the agent already
has one in hand, it would likely elect to take the shorter plan
and just hurl said taco.

10

3/28/15

Dierences
The planner diverges from the FSM and BT in that
it isnt specically hand-authored
Therein lies the dierence in plannersthey actually
solve situaBons based on what is available to do and
how those available acBons can be chained together
One of the benets of this sort of structure is that it
can oVen come up with soluBons to novel situaBons
that the designer or programmer didnt necessarily
account for and handle directly in code

Advantages and Disadvantages

From an implementaBon standpoint, a major plus of the planner is that a

new acBon can be dropped into the game and the planner architecture will
know how to use it
This speeds up development Bme markedly
All the author says is, here are the potenBal things you could do go forth and do
things

Of course, a drawback of this is that authorial control is diminished.

In a FSM or BT, creaBve, outside the box soluBons were the excepBon from the
predictable, hand-authored systems
In a planner, the scripted, predictable moments are the excepBon; you must
specically override or trick the planning system to say, no I really want you to do
this exact thing at this moment

11

3/28/15

Example
A more recent avor of planner is the
hierarchical task network (or HTN) planner
such as was used to great eect in Guerillas
Killzone 2

hLp://aigamedev.com/open/coverage/htn-planning-discussion/

uBlity-based method
Instead of assembling a plan like
the planner, however, the
uBlity-based system simply
selects the single next acBon
All the ingredients are in the
mix and available at all Bmes.
However, you simply select
what it is that would like to
poke at and execute.
You can select it based on what
you have a taste for or what is
most accessible at the moment.

12

3/28/15

uBlity-based method

The progression of AI architectures

throughout The Sims franchise is well
documented
Each potenBal acBon in the game is
scored based on a combinaBon of an
agents current needs and the ability of
that acBon or item to saBsfy that need
The agent then uses an approach
common in uBlity-based methods and
constructs a weighted sum of the
consideraBons to determine which acBon
is the best at that moment
The acBon with the highest score wins

uBlity-based method
While uBlity-based systems can be used in many types of
games, they are more appropriate in situaBons where there
are a large number of potenBally compeBng acBons the AI
can takeoVen with no obvious right answer
In those Bmes, the mathemaBcal approach that uBlity-based
systems employ is necessary to ferret out what the most
reasonable acBon to take is
Aside from The Sims, other common areas where uBlity-
based systems are appropriate are in RPGs, RTS, and
simulaBons.

13

3/28/15

ImplementaBon

Like behavior trees and planners, the uBlity-based AI code is a reasoner

Once an acBon is decided up, the agent sBll must transiBon to a state
The uBlity system simply is selecBng what state to go to next
In the same way, then, just like those other systems, the reasoning code is
all in a single place
This makes building, ediBng, tuning and tweaking the system much more
compartmentalized
Also, like a planner, adding acBons to the system is fairly straight forward
By simply adding the acBon with the appropriate weights, the AI will
automaBcally take it into account and being using it in relevant situaBons
This is one of the reasons that games such as The Sims were as expandable
as they werethe agents simply included any new object into their
decision system without any changes to the underlying code.

Drawback

There isnt always a good way to intuit what will happen in a given situaBon
In a behavior tree, for example, it is a relaBvely simple exercise to traverse the tree
and nd the branches and nodes that would be acBve in a parBcular situaBon

Because a uBlity system is inherently more fuzzy than binary, determining

how the acBons stack up is oVen more opaque
Thats not to say that a uBlity-based AI is not controllable or congurable
In fact, uBlity systems oer a deep level of control
The dierence is that rather than telling the system exactly what to do in a
situaBon, the system is providing suggesBons as to what might be a good
idea
In that respect, a uBlity system shares some of the adaptable aspects of
plannersthe AI simply looks at its available opBons and then decides what
is most appropriate.

14

3/28/15

Neural Network (NN)

This is the caveat emptor of the NN-based AI soluBon
As a type of learning AI, neural nets need to be trained with
test or live performance data
At some point you have to wrap up the training and say, This
is what I have
If a designer wanders in, looks over your shoulder and says,
It looks preLy cool, but in [this situaBon] I would like it to do
[this acBon] a liLle a liLle more oVen, theres really nothing
you can do to change it
About all you can do is try to retrain the NN and hope for the
best

Neural Network (NN)

Lack of designer control aVer the
fact
Unfortunately, this tends to
disqualify NNs and other machine-
learning soluBons from
consideraBon in the game AI
environment where that level of
control is not only valuable but
oVen a requirement

15

3/28/15

Wrap up
This has by no means been an exhausBve
treatment of AI architectures
The purpose was simply to expose you to the
opBons that are out there and why you may
or may not want to select each for the
parBcular tastes and needs of your project
To sum up, though, lets go through the
opBons once again

Wrap up

16

3/28/15

Conclusions
there is no one size ts all solu7on to AI
architectures

You also dont have to limit yourself to a single

architecture in a game

Depending on your needs, your team, and your prior

experience, any of the techniques may be The Right
Way for you to organize your AI
As with any technical decision, the secret is to
research what you can, even try a few things out, and
decide what you like

Homework (HFSM)
Michael Booth. 2009. "The AI Systems of LeV 4
Dead." ArBcial Intelligence and InteracBve
Digital Entertainment Conference (Stanford
University).
(On Moodle)

17

3/28/15

Thanks!

18