Skip to main content

Turn perceptions into Prolog facts

A perception is a plain Kotlin object (see Connect agents to an external environment). A Prolog agent turns it into facts in handlesPerceptionEvents. The snippets come from the vacuum-world example.

Build facts with the 2P-Kt API​

Outside plans there is no DSL scope: build terms with the 2P-Kt core types (it.unibo.tuprolog.core):

Fact.of(Struct.of("position", Integer.of(position.x), Integer.of(position.y)))
Fact.of(Struct.of("facing", facing.toAtom()))

fun Enum<*>.toAtom(): Atom = Atom.of(name.lowercase())

Use lowercase atoms: a string starting with an uppercase letter is a variable in the DSL, and Atom.of("Bob") is the quoted atom 'Bob'. Keeping these functions in one file, as the example does in Beliefs.kt, gives one place where Kotlin values become terms, and where the code that reads the beliefs back finds the same names.

Add what is new, remove what is stale​

Diff the perceived facts with the current ones, so that unchanged facts generate no events and beliefs the agent derived itself (like a memory of visited cells) are left alone. The facts a reading replaces are here the beliefs with the same functors:

fun handleVacuumPerception(event: VacuumPerception, beliefs: Collection<PrologBelief>): AgentUpdate<*> {
    val perceived = event.toBeliefs()
    val previous = beliefs.filter { it.head.functor in perceived.map { new -> new.head.functor } }.toSet()
    // ... plus the remembered dust_at(X, Y) beliefs, see below
    return AgentUpdate.Belief(additions = perceived - previous, removals = previous - perceived)
}

To select the replaced beliefs more finely, match them against queries built with newContextBeliefQuery { } (from it.unibo.jakta.dsl.belief): belief.matchBelief(newContextBeliefQuery { "position"(X, Y) }) != null.

The handler can also keep a memory derived from what it perceives. The vacuum robot adds dust_at(X, Y) for every square it sees dusty and removes it for every square it sees empty, so the memory is always computed from the position the reading was taken at:

val dustSeen = event.cellsWith(Content.DUST).map { cellBelief("dust_at", it) }.toSet() - beliefs.toSet()
val dustGone = event.cellsWith(Content.EMPTY).map { cellBelief("dust_at", it) }.toSet() intersect beliefs.toSet()
return AgentUpdate.Belief(
    additions = perceived - previous + dustSeen,
    removals = previous - perceived + dustGone,
)

Hook it into the agent​

handlesPerceptionEvents { if (it is VacuumPerception) handleVacuumPerception(it, beliefs) else null }

Returning null ignores a perception. Plans then react to the facts as to any belief, e.g. adding.belief { matchingBelief { "position"(X, Y) } }.

The tictactoe example does the same with the board (handleBoardPerception in TicTacToeSkills.kt).