[工程科技]Modeling Knowledge and Deliberation in the Situation Calculus.pdf

Modeling Knowledge and Deliberation in the Situation Calculus ´ Yves Lesperance Department of Computer Science York University Toronto, Canada Ph.D. Seminar, Univ. of Rome, May 2002 Motivation To do planning in IndiGolog, user provides non-deterministic program/plan skeleton, which is put in a “search block”. Interpreter must find plan/strategy that ensures successful execution of search block. This is a very general version of the problem of planning under incom- plete information with sensing actions. Need specification. When is a plan a solution? Can’t be just any program that achieves goal or reaches a final situa- tion of the user supplied program. Should not require deliberation to execute. Often, specifications require planner to return a program of a syntacti- cally restricted form. 1 Our Approach To Formalizing Planning/Deliberation Have developed account of planning under incomplete information that is more abstract, semantically-motivated. Requires planner to return an epistemically feasible program, i.e., a program for which the agent always knows what step to take next no matter how sensing turns out. Account framed as semantics for IndiGolog search/deliberation opera- tor. To characterize epistemically feasible plans, we will use an explicit model of knowledge within the action theory. Before going over our semantics of search, we will look at how knowl- edge is modeled in logic, and in particular in the situation calculus. 2 E.g. Getting on Flight at Airport [Levesque 96] Agent wants to board flight. Does not know gate in advance; mu