Personalising Medication & Activity Regimes Using Novel State Progression Models for Forward Search with PDDL+

Student thesis: Doctoral ThesisDoctor of Philosophy


My work explores the possibility of modelling complex medication instructions (partic- ularly painkillers), as hybrid models and using artificial intelligence planning to personalise dose schedules to suit user preferences such as work hours, sleep schedules and meal times. This is done by modelling the action of medication consumption, as well as using piecewise linear approximation to model the exponential decay of drugs in Pddl. Some existing research already looks at modelling medication consumption schedules with user preferences but few, if any, clearly specify how this is achieved.

On the other hand, planners are becoming increasingly capable of reasoning with continuous dynamics (drug half-lives are a good example of this), as expressed in Pddl+, making them capable of attempting complex real-world problems. That being said, Pddl+ domains with multiple processes that have continuous numeric effects tend to struggle due to the lack of efficient applicability models that utilise the information in states during search to make better search decisions surrounding Pddl+ processes. applicability model to determine the priority or order of processes.

My research consists of three stages: in the first, I model a personalised medicine domain using Pddl+ and devise a linearise-validate cycle to allow us to create medication regimes; in the second, I improve our domain to support activity planning, as well as tweak the planner to support this; and in the final stage, I present an improved state progression model for forward search with Pddl+ alongside two Pddl+ benchmark domains on which I demonstrate the impact of these progression models.
Date of Award1 Jul 2022
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorAndrew Coles (Supervisor), Sanjay Modgil (Supervisor) & Steffen Zschaler (Supervisor)

Cite this