Conditional Partial Plans for Rational Situated Agents Capable of Deductive Reasoning and Inductive Learning

Abstract: Rational, autonomous agents that are able to achieve their goals in dynamic, partially observable environments are the ultimate dream of Artificial Intelligence research since its beginning. The goal of this PhD thesis is to propose, develop and evaluate a framework well suited for creating intelligent agents that would be able to learn from experience, thus becoming more efficient at solving their tasks. We aim to create an agent able to function in adverse environments that it only partially understands. We are convinced that symbolic knowledge representations are the best way to achieve such versatility. In order to balance deliberation and acting, our agent needs to be emph{time-aware}, i.e. it needs to have the means to estimate its own reasoning and acting time. One of the crucial challenges is to ensure smooth interactions between the agent's internal reasoning mechanism and the learning system used to improve its behaviour. In order to address it, our agent will create several different conditional partial plans and reason about the potential usefulness of each one. Moreover it will generalise whatever experience it gathers and use it when solving subsequent, similar, problem instances. In this thesis we present on the conceptual level an architecture for rational agents, as well as implementation-based experimental results confirming that a successful lifelong learning of an autonomous artificial agent can be achieved using it.