AI Techniques
Boids Flocking
Boids flocking is a technique which used to create crowd
behaviour, such as a school of fish. In (mick maus, 2015)’s video titled “How do Boids Work? A Flocking Simulator”
he explains the three guidelines that the boids follow. These are “alignment” which is a rule to determine
the direction at which the boids are travelling. They all follow a common
heading. If one boid changes direction, then all the other boids which change
their alignment based on that. The second rule is “cohesion”, a rule that is applied to the area around the boids.
Each boids will search the area around it, in its “radius” and will move to the middle. This helps to keep all the
boids closer together, even if they are told not to crowd with each other. This
leads me onto the third rule “crowding”,
which basically keeps the boids together, hence the term ‘flocking’ (which
refers to a flock of birds).
The application of boids flocking can be used to form
units of agents, allowing them to travel as one unit. Let’s imagine a scenario
where the player has just set of an alarm, and reinforcements are inbound.
Enemies might form a group, a ‘flock’, when running in towards the player.
There behaviour could be linked, such a not spacing out too far from each
other.
Tactics and Strategy
The best AI from games that we know and love are employed
by tactic and strategy. I’m not just talking from an in-game aspect, but also
from a design aspect. There are certain strategies that are used to make a good
AI system that seem realistic enough that we also forget that it isn’t human. In
(Mark Brown, 2017)’s video titled “What
Makes Good AI?” he highlights some of the aspects of what makes a good AI
such as how AI “Lets the player cheat.”
He explains that in the game FarCry
there is a limit on the number of enemies that can be shooting at the player at
any given time. This means that the player won’t die too quickly, especially if
swarmed by many enemies. This is just one of the ways that AI cooperate with
each other behind the scenes -in a way that most of us players wouldn’t even
realise was a factor.
Swarms of dangerous mobs (unlike the real-world term, the
gaming term for mob refers to just one entity) can make the player feel danger.
But what makes a truly good AI are those that interact with them and the environment
around them. (Champandard, 2007) outlines a couple of “behaviours” that I believe are necessary
to AI, these are:
1.
“Assuming
a intermediate position to search for cover, e.g. like dropping down
on one knee.
2.
Playing a short animation while looking for
cover, and signaling the threat to team mates.”
When these
actions are performed by AI it can appear, even if they have their own thought
process, that they are moving as a unit.
In the video that I outlined earlier, (Mark Brown, 2017) explains that
the body language of an AI is important to us as players. These act as cues
that we can use to know what our actions mean to them, and how we can interact
with the environment. For example, if an AI hears footsteps, the agent might
turn around and walk towards where the sound came from. This was a principle
that was used in the franchise Metal Gear
Solid. In (YongYea, 2014)’s video, he demonstrates the effect that the
player character can have by throwing magazines and knocking on walls. This
makes the AI appear intelligent, even if implementing something like this is a
basic technique.
Behaviour trees and FSMs can play a part in how AI
behaves. (Tangness, 2014) explains in his online article about
the game Splinter Cell: Blacklist how
AI agents behave using states. He claims that there is are four states, these
being: “Normal patrol”, “Normal Investigate”, “Heightened Mode” and “Combat Mode”. He explains how the AI
moves through these different states based on different factors such as whether
the player has made a sound (which they are extremely sensitive to), or has
been detected. These behavioural trees make the guards seems intelligent AI
agents, and allow for a more immersive experience.
Hacks
Hacks in video games refer to the AI having knowledge
that shouldn’t be openly available to them, such as constantly knowing the
player’s state, i.e. when they’re out of ammo, low on health etc. It ruins the
feel of the game, especially if they are meant to be realistic. No one likes to
feel like every move they make is being predicted. It makes the player feel
like the AI is in some way cheating; this is because it basically is. (tvtropes, n.d.) outlines the problem
of what he calls an “All Seeing AI”,
he explains that “Stealth
is useless, no surprises are possible, and it will (almost) never miss a shot.
Consequently, players should not bother with misdirection, flanking, or other
forms of deception and psychological warfare that would work wonderfully
against actual humans.” When we play games, we want to feel that we are
smarter than who we are up against, it’s part of the reason that makes computer
games so competitive. It is however, a terrible position to put the player in
when you are leaving them with no power over the flow of information that the
AI receives. It’s frustrating when the AI lands every shot without fail. All of
this breaks the immersion of the game itself. How are we meant to feel like the
game is pulling us in when it’s constantly undermining the aspects of reality?
Hacks aren’t all bad though,
they can be used to determine when a cutscene will occur -something that is
often appreciated by the player. The game can use the player’s location to open
the world up or spawn a boss.
