Using a POMDP controller to guide persons with dementia through activities of daily living

Researchers at the Centre for Studies in Aging and at Simon Fraser University are developing ubiquitous assistive technology to aid persons with dementia complete routine activities. To ensure that the system is useful, effective, and safe, it must be able to adapt to the user and guide him/her in an environment that may not be fully observable. This paper discusses the merits of using partially observable Markov decision process (POMDP) algorithms to model this problem as POMDPs are able to provide robust and autonomous control under conditions of uncertainty. A POMDP controller is being designed for the current prototype, which guides the user through the activity of handwashing.