作者: YI WANG , SHIQI ZHANG , JOOHYUNG LEE
DOI: 10.1017/S1471068419000371
关键词:
摘要: To be responsive to dynamically changing real-world environments, an intelligent agent needs perform complex sequential decision-making tasks that are often guided by commonsense knowledge. The previous work on this line of research led the framework called "interleaved reasoning and probabilistic planning" (icorpp), which used P-log for representing commmonsense knowledge Markov Decision Processes (MDPs) or Partially Observable MDPs (POMDPs) planning under uncertainty. A main limitation icorpp is its implementation requires non-trivial engineering efforts bridge formalisms. In paper, we present a unified integrate icorpp's components. particular, extend action language pBC+ express utility, belief states, observation as in POMDP models. Inheriting advantages languages, new provides elaboration tolerant representation reflects idea design system pbcplus2pomdp, compiles description into model can directly processed off-the-shelf solvers compute optimal policy description. Our experiments show it retains while avoiding manual bridging reasoner planner.