Planning with Multiple Knowledgebases

[AdamSobieski]

Operators could utilize multiple knowledgebases with respect to their preconditions and effects.

Consider that the methods used in operators' preconditions and effects have an implicit knowledgebase argument, e.g., ?world. This parameter could be made explicit so that the knowledgebase argument could be varied.

holds(?world, expression);

add(?world, expression);
delete(?world, expression);

The modeling and definition of operators and of their preconditions and effects would be impacted by the capability for methods like holds, add, and delete to work with multiple knowledgebases.

Reasoning about and planning with multiple knowledgebases could, for instance, allow operators to their describe preconditions and effects involving both a system's environment and the contents of an utterance to be generated. This is depicted in the following example of a precondition which utilizes multiple knowledgebases:

holds(?world, expression1);
holds(?orator, expression2);
holds(?utterance, expression3);
holds(?audience, expression4);

Development and debugging scenarios could be enhanced by having the capability to use multiple knowledgebases. Developers could inspect or query the contents of multiple small, single-purpose knowledgebases instead of having to interact with one large, multi-purpose knowledgebase.

Any thoughts on first-class expressions, first-class knowledgebases, and these ideas that operators' preconditions and effects could utilize multiple knowledgebases?

[AdamSobieski]

Here is a sketch of an operator definition which utilizes multiple knowledgebases in its preconditions. The first knowledgebase, ?world, could be available as a keyword, like the ?this keyword. The second knowledgebase, ?agent1_intention, is obtained from ?agent1 and then also utilized in the preconditions.

public operator example(?agent1)
{
    preconditions
    {
        builtin:holds(?world, {
            rdf:type(?agent1, agents:Agent).
            agents:intention(?agent1, ?agent1_intention).
            rdf:type(?agent1_intention, builtin:Knowledgebase).
        }).
        builtin:holds(?agent1_intention, { ex:expr(ex:x, ex:y). }).
    }
    effects
    {
        // thinking about syntax for modifying both knowledgebases, '?world' and '?agent1_intention', here
    }
}

In this example, there is a ?world knowledgebase, and also every agent (of type agents:Agent) has a knowledgebase or graph attached, via a agents:intention property, which can be used as another knowledgebase in operators' preconditions. This is, in my opinion, better than storing each agent's beliefs, desires, and intentions in one large ?world knowledgebase and having to use more complex expressions to do so. This expressiveness resembles Semantic Web datasets which contain named graphs (see: TriG).

public operator example2(?kb_obj)
{
    preconditions
    {
        builtin:holds(?world, { rdf:type(?kb_obj, builtin:Knowledgebase). }).
        builtin:holds(?kb_obj, { ... }).
    }
    effects
    {
        // would be able to access both '?world' and '?kb_obj' here
    }
}

Hopefully, support for planning with multiple knowledgebases will be useful.