CO2301 Games Development 1 Week 2 Finite State Machines Transitions

1
CO2301 - Games Development 1Week 2Finite State
Machines Transitions

• Gareth Bellaby

2

• Finite State Machines

3
Finite State Machines

• A FSM is a machine which models states,
  transitions between states, and actions.
• State Transition Diagram.

4
Convention being used

• Box represents a state.
• Line represents a transition between states,
  single-direction only.
• An arrow with a dot at the end indicates the
  start state.
• A state can just be an start state (nothing leads
  into the state).
• A state can just be an end state (nothing leads
  from the state).

5
FPS Monster

• Simple monster that reacts when the player comes
  into range and is in line-of-sight. Note
  different convention.

Attacking
No hit points
Attacks
Idle
Start
Dying
6
FPS enemy
Start
See Enemy
Attack
No Enemy
Wander
Low Health
Flee
Taken from Intro to Game Development
No Enemy
7
UML

• UML lends itself well to the design of a FSM.

8
Characteristics

• Two type of machine
• Actions occur within a state
• Actions occur during a transition
• In the games industry the distinctions between
  the two types of FSMs are blurred.
• It is common to see a FSM in which actions occur
  within a state and in which actions occur during
  a transition.

9
Simple Strategic RTS
Build (middle game)
Resource Level
Start
Numerical Inferiority
Explore (early game)
Numerical Superiority
Attack
10
Thief (approximately...)
Suspicious Sound/Sight
Suspicious Sound/Sight
Minor Sound
Suspicious
Definite Sound/Sight
Timer
Not alert
Definite Sound/Sight
Fully Alerted
Start
Suspicious Sound/Sight
Timer
11
Pac-Man Ghost (simplified)
Power Pill
Hunted
Power Pill
Eaten
Timer
Hunter
Resurrect
Start
Eaten
12
Pac-Man Pac-Man (simplified)
Power Pill
Power Pill
Hunter
Timer
Hunted
Collision
Start
Dying
If lives gt 0
13
Capabilities

• A game actor may have more than one FSM. One FMS
  could be swapped out and replaced by another.
• A FSM could cause a cause in another FSM, e.g. a
  guard alerting another guard.
• Probabilities
• Randomness
• Extra information
• Extra actions

14
Characteristics of FSMs

• To summarise, FSMs are
• Flexible
• Appropriate
• Comprehensible and straightforward
• Can become unwieldy especially
• if FSM is piled upon FSM
• lots of interconnecting FSMs (feedback)
• Additions, special cases, etc. undermine the
  structure.

15

• Transitions

16
Transitions

• Consider a threshold from one state to another.
• Imagine a guard in a FPS. All of the following
  could give rise to an activation value
• Seeing something unusual
• Hearing something unusual
• Hearing something specific like a cry
• Receiving an instruction from another guard
• Seeing a guard who has been hurt
• Seeing a guard being hurt

17
Transitions

• A simple sensory model could be implemented using
  a set of rules, e.g. a set of if... then
  statements. Such a model may work, but it is
  simple.
• I want to capture something more complex
• A definite sighting of the player alerts the
  guard, as does a loud noise.
• A possible sighting of the player or a quiet
  noise just makes the guard suspicious.
• But... a possible sighting of the player combined
  with a quiet noise alerts the guard.
• And... repeated quiets noise occurring over a
  short period of time will alert the guard.
• But... over time a suspicious guard will get less
  suspicious if nothing else happens.

18
Transitions
This is the same method as that used in neural
nets
Sum the activation values.

• input signals xi
• weights wi
• activation level ?xiwi
• threshold function f

19
Transitions

• For example, imagine a Thief type game. One input
  could be sight and we have an activation input
  based on how visible the player is.
• A second input could be sound and we have another
  activation input based on how loud the player is.
• The two inputs are summed together. If a
  particular threshold is reached then the guard is
  alerted.
• The important thing to note is that vision or
  sound can cause activation on their own, but the
  combination of the two increases the likelihood
  of activation.
• Now imagine extending this model using, for
  example, activation combined with degradation
  over time... leads to a sophisticated model.