GOLOG sounds similar to your idea of encoding actions in the logic.
There are 2 types of terms: fluents and actions.
Examples of fluents are:
Fluents evaluates to true or false. For example At(room1) = true.
Examples of actions are:
And a GOLOG program can consist of these statements:
These statements can be combined with regular Prolog statements to create a new type of programming, but I'm not sure if it includes automated planning. (Will update later when I finish the lecture notes, Action Programming Languages by Thielscher, 2008).
YKY,
ReplyDeleteYes, this bears some similarity to my proposal... this general scheme is really quite simple, and has sprung up a number of times, but with interesting differences.
I don't see a strong similarity. But I don't know how the little "WHILE language" you cite is integrated with the Prolog part (and the "IO monad" ;-) )
ReplyDeleteMy latest thinking is that inference should NOT be mixed with planning, and the KB should represent actions explicitly, ie, actions are just terms that can be reasoned about; inference rules should NOT directly trigger actions.
ReplyDeleteThis way, the inference engine is about PURE reasoning.
Why I opt for this approach, is the observation that planning is just a special way of answering "how" questions. When Genifer asks "how can I (Genifer) achieve this goal?" she is performing planning; but Genifer can also ask, eg: "how can the monkey get the banana?" (ie, planning for others).
Genifer can even ask "how did the American economy get to be like this?" which is not really a planning problem, but the mechanism to search for the answer is a generalization of planning.