Incorporating Belief Change in a Theory of Action

The overall goal of this work is to try to bring together work in belief change, as represented by the AGM approach (Alchourrón, Gärdenfors, and Makinson 1985), and work in reasoning about action, specifically the situation calculus (Reiter 2001). On the one hand, the AGM approach has been around for a while, and provides the dominant framework for belief change; however there has been little done in successfully applying this approach in realistic situations or with respect to applications. By locating the AGM approach in the situation calculus one can take into account the fact that the world is dynamic, that it evolves as the result of physical and sensing actions, and the like. On the other hand, the (basic approach to the) situation calculus assumes a great deal about the domain: that actions always work as intended, that unforeseen events don’t arise, that sensors work, etc. Clearly, actions may sometimes fail or have unexpected outcomes, the agent must deal with possibly incorrect information, sensors may not work as advertised, etc. So the goal here is to try to come up with a qualitative theory of an agent in a dynamic world where things may not go as expected, where the agent may learn new information by being told something or by sensing, and where the agent must maintain its stock of beliefs as best it can. In the case of the situation calculus there has been substantial work on generalising the original formalism. The proposed work builds on (Delgrande and Levesque 2012; 2018); this work in turn derives from (Shapiro et al. 2000) (which in turn generalises (Scherl and Levesque 2003)). Other related work includes, notably, (Fang and Liu 2013; Schwering, Lakemeyer, and Pagnucco 2017).