Handle logic rules

[meneguzzi]

As a feature request, I'd like for the planner to handle logic rules, as shown in the healthcare domain.

[Maumagnaguagno]

Some details about the healthcare domain that we need to consider:

• Some axioms are redefined to an equal (null) or different description.
  • If each definition adds one possible case the evaluation is unnecessary for repeated cases.
• Some axioms and preconditions contain a complex expression.
  • Today only conjunctions are supported, disjunctions are other branches of the search.
• More than one problem defined in the problem file.
  • Can it be broken or it has an special meaning?

It will take some time to solve each detail for the automatic converter to handle it.

Some parts of the domain are confusing:

• The operator suspend is redefined, is it correct?
• Why is ?t between parentheses in some methods (:method (achieveGoal g4 ?gi (?t) c) ...

[meneguzzi]

• The complex expressions are useful, so I'd prefer if they actually worked. You can probably compile them away when you translate things to ruby. If you translate the condition into a disjunctive normal form, you could change Hypertension.applicable? to process lists of lists of the disjunctions.
• I see a problem now with suspend, there should be one suspend for goals and another for commitments, thanks for pointing this out
• the ?t in parenthesis is to enforce it to the a list (so parameters now can be lists too)

[Maumagnaguagno]

I already started to model expressions to replace applicable?.
The goal is to eventually support and or not call forall exists assign in the intermediate representation to be able to convert the parsed data as needed.

Unification of lists is currently not supported, every term is internally represented as a String.

UPDATE:

• Since commit 58eb7a2 operator preconditions are handled by an AST evaluator (compute).
• Since commit 278e4f4 operator preconditions are evaluated by Ruby logical connectors.
• Since commit db737ad operator preconditions and axioms may contain axioms.

[Maumagnaguagno]

Now we lack the DNF, call (used 30 times for equality tests) and assign (used 2 times).
I will focus on supporting call and then DNF.

Questions:

• I am not familiar with JSHOP assignment scope, is assign really required, couldn't you just reuse the variables?
• I also did not found eqGSCP definition.

(:method (achieveGoal ?g ?gi ?a ?gv)
  genericEnticeToAchieve
  (
    (activeG ?g ?gi ?gv)
    (goal ?g ?gi ?a)
    (commitment ?c ?ci ?a ?d)
    (assign ?cv ?gv)
    (eqGSCP ?g ?gv ?c ?cv)
    (goal ?g2 ?gi2 ?a)
    (assign ?gv2 ?cv)
    (eqGSCQ ?g2 ?gv2 ?c ?cv)
    (call != ?g ?g2)
  )
  (
    (entice ?g ?gi ?gv ?c ?ci ?cv ?a ?d)
    (detach ?c ?ci ?cv)
    (deliver ?g2 ?gi2 ?gv2 ?c ?ci ?cv ?a ?d)
    (achieveGoal ?g2 ?gi2 ?gv2 ?a)
  )
)

[meneguzzi]

I don't seem to use assign in my code for the healthcare scenario, so I think it's OK not to have it implemented.

As for the eqGSCP, it seems this is missing there, but the definition of all the axioms related to reasoning patterns are as follows

;==========================================================
  ;Equalities between conditions on commitments and on goals
  ;==========================================================
  ; True if the success condition of ?g matches the antecedent of ?c 
  ;(?gv and ?cv will unify with the specific instance of ?g and ?c that match this)
  (:- (eqGSCP ?g ?gv ?c ?cv) (and (imply (s ?g ?gi ?gv) (p ?c ?ci ?cv)) (imply (p ?c ?ci ?cv) (s ?g ?gi ?gv)) ) ) ; Basically a logical equivalence <-> 
      ;TODO make sure that the variables ?gv and ?cv get instantiated
  ; True of the success condition of ?g matches the consequent of ?c
  (:- (eqGSCQ ?g ?gv ?c ?cv) (and (imply (s ?g ?gi ?gv) (q ?c ?ci ?cv)) (imply (q ?c ?ci ?cv) (s ?g ?gi ?gv)) ) ) ; Basically a logical equivalence <->
  ; True of the the antecedent of commitment ?c1 matches the consequent of commitment ?c2 (so they are reciprocal)
  (:- (eqCPCQ ?c1 ?cv1 ?c2 ?cv2) (and (imply (p ?c1 ?ci1 ?cv1) (q ?c2 ?ci2 ?cv2)) (imply (q ?c2 ?ci2 ?cv2) (p ?c1 ?ci1 ?cv1)) ) ) ; Basically a logical equivalence <->

[Maumagnaguagno]

There are assign statements here

[meneguzzi]

I can live without them, those were just testing code.

[meneguzzi]

@Maumagnaguagno, I was trying to parse my domain again and find the replications, but one parse error you seem to report is a "redefinition" of the axioms. However, when you have axioms for the p and q of different commitments, you are talking about different possibilities for the variables in the axiom (like when you have multiple rules in Prolog).
For example, the code from here
to here, must work

We need to support this "redefinition", much in the same way we can have different methods for the same task. While I agree that redefining actions may create ambiguity, there is no such problem for the axioms.

[Maumagnaguagno]

Yes, I am working on it.

[Maumagnaguagno]

Let's try to deal with the constants first then. Check the boxes here when you are finished:

• Handling constants in axiom declarations
• Support calls/axioms/attachments in methods
• Propagate ground lists
• Unify list terms
• Support disjunctions (DNF) in methods (complex unification)