TITLE: Policy learning for hybrid systems under uncertainty
SPEAKER: Daniele Magazzeni (Department of Informatics, King's College London)

ABSTRACT:

Real-world systems often present a mixed discrete-continuous dynamics, and operate in physical situations characterised by a high level of uncertainty. Planning under uncertainty and with hybrid domains is one of major challenges for the AI Planning community. Recently, a new approach called -Y´plan-based policy learning¡ has been shown to be a powerful technique for obtaining robust intelligent behaviour in the face of uncertainty. Indeed, a robust behaviour can often be learned by sampling a large number of cases, planning what to do precisely in those cases and then using machine learning techniques to learn a general policy for what to do in future unseen cases. This talk will present two examples of this approach applied to the problem of load management for multiple batteries and the problem of tracking a moving patch of water in the ocean. The talk will also outline some open issues in modelling and solving hybrid planning domains, which represent interesting problems also for other areas, such as formal methods, model checking, constraint programming.