State Machine and the Abstractions
Hello, For the purposes of a simple strategy game, I'd like to build an EDSL that expresses missions. A mission could be represented as a state machine. With basic bricks such as actions (MoveTo, ShootAt...) or tests (EnemiesAround, LowHealth...), I could (ideally dynamically) build some strategic behaviors for the units. I will take the example of a patrol. Applied to a unit (or a group of units), it dictates : go from point 1 to point 2 and then go back and repeat. But when you detect an enemy near, leave the patrol path, destroy it and then resume your patrol where you left it. So if I consider my mission as a monad: data Mission = MoveTo Point | ShootAt Unit patrol = do MoveTo point1 MoveTo point2 patrol So far so good, but there, the only advantage to use a monad instead of a list of MoveTo's is the do-notation. And I lack the expression of tests. Using a GADT it could be: data Mission a where MoveTo :: Point -> Mission () ShootAt :: Unit -> Mission Bool -- If we have destroyed it or not EnemiesAround :: Mission [Unit] -- The enemies that are maybe in sight LowHealth :: Mission Bool -- If I should retreat ... -- (Monad Mission could be nicely expressed using Heinrich Apfelmus' * operational* package) patrol = do MoveTo point1 MoveTo point2 enemies <- EnemiesAround mapM_ ShootAt enemies patrol Aaaaaaaand... here comes the trouble: the actions are done *sequentially*. My units will move and then look at enemies, they will not monitor their environment while they move. So I need a way to say: A is your action of patrolling. B is your action of surveillance. Do both in parallel, but B is preponderant, as if it successes (if enemies are there) it takes over A. So, it is as if I was running two state machines in parallel. Moreover, the last line (the recursive call to patrol) is wrong, as it will restart the patrol from the beginning, and not from where it has been left. But this could be corrected by addind a test like "which point is the closest". So I thought about Arrows, as they can express sequential and parallel actions, but I don't know if it would be a right way to model the interruptions/recoveries. What do you think about it? Do you know of similar situations and of the way they've been solved?
On Fri, May 27, 2011 at 2:06 PM, Yves Parès
Hello,
For the purposes of a simple strategy game, I'd like to build an EDSL that expresses missions. A mission could be represented as a state machine. With basic bricks such as actions (MoveTo, ShootAt...) or tests (EnemiesAround, LowHealth...), I could (ideally dynamically) build some strategic behaviors for the units. I will take the example of a patrol. Applied to a unit (or a group of units), it dictates : go from point 1 to point 2 and then go back and repeat. But when you detect an enemy near, leave the patrol path, destroy it and then resume your patrol where you left it.
So if I consider my mission as a monad: data Mission = MoveTo Point | ShootAt Unit
patrol = do MoveTo point1 MoveTo point2 patrol
So far so good, but there, the only advantage to use a monad instead of a list of MoveTo's is the do-notation. And I lack the expression of tests. Using a GADT it could be:
data Mission a where MoveTo :: Point -> Mission () ShootAt :: Unit -> Mission Bool -- If we have destroyed it or not EnemiesAround :: Mission [Unit] -- The enemies that are maybe in sight LowHealth :: Mission Bool -- If I should retreat ...
-- (Monad Mission could be nicely expressed using Heinrich Apfelmus' operational package)
patrol = do MoveTo point1 MoveTo point2 enemies <- EnemiesAround mapM_ ShootAt enemies patrol
Aaaaaaaand... here comes the trouble: the actions are done sequentially. My units will move and then look at enemies, they will not monitor their environment while they move. So I need a way to say: A is your action of patrolling. B is your action of surveillance. Do both in parallel, but B is preponderant, as if it successes (if enemies are there) it takes over A. So, it is as if I was running two state machines in parallel. Moreover, the last line (the recursive call to patrol) is wrong, as it will restart the patrol from the beginning, and not from where it has been left. But this could be corrected by addind a test like "which point is the closest".
Could this be expressed using a new verb in your language?
data Mission a where MoveTo :: Point -> Mission () ShootAt :: Unit -> Mission Bool -- If we have destroyed it or not EnemiesAround :: Mission [Unit] -- The enemies that are maybe in sight LowHealth :: Mission Bool -- If I should retreat . . . Tasks :: [Mission ()] -> Mission () -- goals to be achieved concurrently Options :: [Mission ()] -> Mission () -- pick one of these
You'd then need to analysis and interpretation tools to correctly do the right thing with it. If the game is state-machine driven, you would need sub state-machines for each possibility under 'Tasks' or 'Options'. Antoine
So I thought about Arrows, as they can express sequential and parallel actions, but I don't know if it would be a right way to model the interruptions/recoveries. What do you think about it? Do you know of similar situations and of the way they've been solved?
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
On 27 May 2011 20:06, Yves Parès
So I thought about Arrows, as they can express sequential and parallel actions, but I don't know if it would be a right way to model the interruptions/recoveries. What do you think about it? Do you know of similar situations and of the way they've been solved?
Resumption monads? Take a look at William Harrison's work especially the CellSys DSL and the models of operating systems: http://people.cs.missouri.edu/~harrisonwl/publications.html
I suggest you take a look at MonadPrompt and/or Operational (two competing
packages, one of which I wrote).
And yes, you probably need some operation
Concurrent :: [Mission ()] -> Mission ()
or
Interrupt :: Mission () -> Mission Bool -> Mission () -> Mission ()
which runs its first argument until the second argument returns True, then
switches.
On Fri, May 27, 2011 at 12:06 PM, Yves Parès
Hello,
For the purposes of a simple strategy game, I'd like to build an EDSL that expresses missions. A mission could be represented as a state machine. With basic bricks such as actions (MoveTo, ShootAt...) or tests (EnemiesAround, LowHealth...), I could (ideally dynamically) build some strategic behaviors for the units. I will take the example of a patrol. Applied to a unit (or a group of units), it dictates : go from point 1 to point 2 and then go back and repeat. But when you detect an enemy near, leave the patrol path, destroy it and then resume your patrol where you left it.
So if I consider my mission as a monad: data Mission = MoveTo Point | ShootAt Unit
patrol = do MoveTo point1 MoveTo point2 patrol
So far so good, but there, the only advantage to use a monad instead of a list of MoveTo's is the do-notation. And I lack the expression of tests. Using a GADT it could be:
data Mission a where MoveTo :: Point -> Mission () ShootAt :: Unit -> Mission Bool -- If we have destroyed it or not EnemiesAround :: Mission [Unit] -- The enemies that are maybe in sight LowHealth :: Mission Bool -- If I should retreat ...
-- (Monad Mission could be nicely expressed using Heinrich Apfelmus' * operational* package)
patrol = do MoveTo point1 MoveTo point2 enemies <- EnemiesAround mapM_ ShootAt enemies patrol
Aaaaaaaand... here comes the trouble: the actions are done *sequentially*. My units will move and then look at enemies, they will not monitor their environment while they move. So I need a way to say: A is your action of patrolling. B is your action of surveillance. Do both in parallel, but B is preponderant, as if it successes (if enemies are there) it takes over A. So, it is as if I was running two state machines in parallel. Moreover, the last line (the recursive call to patrol) is wrong, as it will restart the patrol from the beginning, and not from where it has been left. But this could be corrected by addind a test like "which point is the closest".
So I thought about Arrows, as they can express sequential and parallel actions, but I don't know if it would be a right way to model the interruptions/recoveries. What do you think about it? Do you know of similar situations and of the way they've been solved?
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Yves Parès wrote:
For the purposes of a simple strategy game, I'd like to build an EDSL that expresses missions. A mission could be represented as a state machine.
With basic bricks such as actions (MoveTo, ShootAt...) or tests (EnemiesAround, LowHealth...), I could (ideally dynamically) build some strategic behaviors for the units. I will take the example of a patrol. Applied to a unit (or a group of units), it dictates : go from point 1 to point 2 and then go back and repeat. But when you detect an enemy near, leave the patrol path, destroy it and then resume your patrol where you left it.
So if I consider my mission as a monad: data Mission = MoveTo Point | ShootAt Unit
patrol = do MoveTo point1 MoveTo point2 patrol
[...] So I need a way to say: A is your action of patrolling. B is your action of surveillance. Do both in parallel, but B is preponderant, as if it successes (if enemies are there) it takes over A. So, it is as if I was running two state machines in parallel.
There are several methods to specify state machines, sequential composition of actions is just one of them. Let me elaborate. First and foremost, you can express every state machine as an automaton. For instance, your example above could be written as state1 --> (MoveTo point1, state2) state2 --> (MoveTo point2, state3) state3 --> ((), state1) An automaton is a set of states and transitions between them, and you should imagine that all your state machines work like this. Of course, while all state machines *work* like this, this does not mean that you have to *specify* them like this. For instance, writing down the states for a long sequence of actions like do moveTo point1 moveTo point2 shoot moveTo point3 ...etc. would be very cumbersome, you'd have to invent one dummy state for each action in the sequence. And this is where do-notation comes in: it's a way to specify long sequences of actions and have the interpreter automatically generate the intermediate dummy states! As you note, however, not all state machines are sequences of actions, far from it, actually. Sometimes, you want to write Flee --> MoveTo point0 Attack --> shoot `during` MoveTo point1 Well, nobody forces you to use do-notation in that case, right? In other words, I propose that you invent a small set of *state machine combinators* that allow you to freely mix do-notation (or something less powerful) with state transitions. Parallel composition corresponds to pairing states. (Chances are that you can express everything with do-notation by introducing threads and combinators like during or fork , but I don't know whether that's the best way to go about it. It's worth trying, but keep in mind that the goal is to have an expressive set of combinators, not to shoehorn everything into monads.) Best regards, Heinrich Apfelmus -- http://apfelmus.nfshost.com
participants (5)
-
Antoine Latter -
Heinrich Apfelmus -
Ryan Ingram -
Stephen Tetley -
Yves Parès