AbstractAutomated planning in hybrid domains is a topic that is gaining significant research interest. Hybrid representation is needed in real-world applications involving both continuous and discrete aspects. However, the hybrid nature of such problems poses a serious challenge to existing planning approaches, as the traditional focus of the research community is on classical domains with discrete actions. State-of-the-art domain-independent planning systems excel in dealing with discrete variables, boolean aspects, and discontinuities, however, reasoning about the continuous numeric change, which is an essential element of a hybrid domain, can be particularly challenging for them. The field of control systems engineering, on the contrary, focuses on modelling and controlling dynamical processes with continuous numeric changes, but struggles in the presence of discrete changes.
In this thesis, we present a framework for domain-independent planning with hybrid domains. It is based upon the idea of combining the complementary strengths of automated planning and control theory for solving PDDL+ hybrid planning problems. The Model Predictive Control technique from the field of control theory has been employed in the planning framework for guiding the forward search process for the continuous part of the hybrid domain. The Model Predictive Control proposes an optimal trajectory, possibly constituted by the system due to continuous processes. It is utilised as a guide for the forward search by the planner, which uses the classical planning techniques for reasoning about the discrete elements of the hybrid domain.
Experimental evaluation, using a large set of problem instances from a range of benchmark domains, demonstrates the effectiveness of the proposed planning approach. Empirical analysis shows that the proposed approach can drastically reduce the planning time and outperform state-of-the-art planning systems, particularly in the cases of domains involving processes with coupled equations and variable rates of change.
|Date of Award||2023|
|Supervisor||Mauro Vallati (Main Supervisor) & Diane Kitchin (Co-Supervisor)|