A General Representation and Approximate Inference Algorithm for Sensing Actions

Abstract

Sensing actions, which allow an agent to increase its knowledge about the environment, are problematic for traditional planning languages. In this paper we propose a very general framework for representing both changes to the real world and to the knowledge of an agent, based on a first order linear time calculus. Our framework is more general than most existing approaches, because our semantics explicitly represents, for each point in time, not only the agent's knowledge about that timepoint, but also about the past and the future. By applying a general approximation method for classical logic to this framework, we obtain an efficient and sound but incomplete reasoning method.

A history based approximate epistemic action theory for efficient postdictive reasoning

“…Their expressiveness and inference capabilities are comparable to PWS approaches, but corresponding implementations can be more efficient in practice [53]. [48] EXP ---Epistemic SC Scherl & Levesque [47] EXP ---AOL Lakemeyer & Levesque [27] ∞ ----Epistemic FC Thielscher [50] EXP --( ) CFF Hoffmann & Brafman [22] EXP ---EFEC Ma et al [31], Miller et al [35] EXP -Approach by Vlaeminck et al [56] EXP ---Approach by Fagin et al [16] EXP --TH-2 -Disjunctive state approaches DECKT Patkos & Plexousakis [40] EXP -( ) PrAO To [54] EXP --- [18] LIN --( ) -TH-3 Approximations of PWS (e.g. [48]).…”

“…Concerning the temporal dimension of knowledge two approaches stand out, namely Ma et al [31] and Vlaeminck et al [56]. Ma et al [31] propose an epistemic extension to the Event Calculus [25] which considers possible world histories instead of possible worlds.…”

“…Concurrent acting and sensing is possible with EFEC, and an implementation exists as answer set programming. 4 Another PWS-based first-order logical framework to model a temporal dimension of knowledge is provided by Vlaeminck et al [56]. The framework employs first-order-reasoning such that it allows for elaboration tolerant postdiction and the projection problem is solvable in polynomial time.…”