Finite-state machines

Finite-state machine is often used for artificial intelligence routines in games. They can model different states character can be: patrolling, searching for food, investigating noise and fighting. There is a small DSL for defining finite-state machines supplied with pyherc.

Sample configuration

Following code is a sample definition for a very simple finite-state machine. It has two states addition and subtraction.

(defstatemachine SimpleAdder [message]
  "finite-state machine for demonstration purposes"

  "add 1 to message, 0 to switch state"
  (addition initial-state
            (active (+ message 1))

            "message 0 will change state"
            (transitions [(= message 0) subtraction]))

  "substract 1 from message, 0 to switch state"
  (subtraction (active (- message 1))

               "message 0 will change state"
               (transitions [(= message 0) addition])))

In order to use the finite-state machine, one needs to create an instance of it and call it like a function:

=> (setv fsm (SimpleAdder))
=> (fsm 1)
2
=> (fsm 2)
3
=> (fsm 0)
-1
=> (fsm 1)
0
=> (fsm 2)
1

As you can see, fsm will first return the argument passed to it plus 1. As soon as 0 is passed in, finite-state machine switches to subtraction state and starts returning the argument passed to it minus 1. Passing a 0 again will change the state back to addition.

Sometimes there’s need to perform extra initialisation when finite-state machine is created or store data across different states. Following example highlights how --init-- and state forms can be used to achieve this.

(defstatemachine Minimal [message]
  "default initializer"
  (--init-- [bonus] (state bonus bonus))
  "handle message"
  (process initial-state
           (active (* message (state bonus)))))

Following example shows how the finite-state machine defined in previous example can be used:

=> (setv fsm (Minimal 3))
=> (fsm 1)
3
=> (fsm 5)
15

As you can see, the parameter supplied during initialization of finite-state machine is stored under symbol bonus and used when finite-state machine is activated.

Syntax of finite-state machine definition

Finite-state machine is defined with (defstatemachine <name> <parameters>) form. <name> defines name of the class that will encapsulate finite-state machine definition. <parameters> is a list of zero or more symbols that define function interface that the finite-state machine will have. Keyword only, optional or other special parameter types are not supported.

Inside of defstatemachine form, there are one or more state definitions. Strings are allowed and they’re treated as comments (ie. ignored). Format of state definition is (<name> [initial-state] [(on-activate ...)] [(active ...)] [(on-deactivate ...)] [(transitions ...)]). <name> is name of the state, it should be unique within a finite-state machine as transitions refer to them. One and only one of the states should be marked as an initial-state. This is the state the finite-state machine will enter when first activated. Rest three forms are all optional. Order of the forms is not significant. Symbols defined in <parameters> block of defstatemachine are available to all of these three functions. Strings are allowed and they are treated as comments (ie. ignored). Special form --init-- can be used to create initializer method for finite-state machine. It has syntax of (--init-- <parameters> <body>). <parameters> is a list of symbols that are to be added in --init-- method of the finite-state machine and <body> is one or more s-expressions that are to be executed when finite-state machine is initialized.

First one is on-activate, which defines code that is executed when the given state is activated. Second one is active which defines code that is executed every time for the active state when finite-state machine is activated. on-activate is mirrored by on-deactivate, which gets executed every time a state deactivates. The last one is transitions. It defines one or more two element lists, where the first element is test and second element is symbol of a state to switch if the test returns true. transitions are checked for the active state every time finite-state machine is activated and it is performed before active code is executed.

In order to store data and pass it between states, state macro can be used. It has syntax of: (state <symbol> [value]). <symbol> is the stored data being accessed. If optional value is supplied, stored data is updated. In any case state returns the current value of the data.