Assistive systems for persons with cognitive disabilities (e.g. dementia) aredifficult to build due to the wide range of different approaches people cantake to accomplishing the same task, and the significant uncertainties thatarise from both the unpredictability of client's behaviours and from noise insensor readings. Partially observable Markov decision process (POMDP) modelshave been used successfully as the reasoning engine behind such assistivesystems for small multi-step tasks such as hand washing. POMDP models are apowerful, yet flexible framework for modelling assistance that can deal withuncertainty and utility. Unfortunately, POMDPs usually require a very labourintensive, manual procedure for their definition and construction. Our previouswork has described a knowledge driven method for automatically generating POMDPactivity recognition and context sensitive prompting systems for complex tasks.We call the resulting POMDP a SNAP (SyNdetic Assistance Process). Thespreadsheet-like result of the analysis does not correspond to the POMDP modeldirectly and the translation to a formal POMDP representation is required. Todate, this translation had to be performed manually by a trained POMDP expert.In this paper, we formalise and automate this translation process using aprobabilistic relational model (PRM) encoded in a relational database. Wedemonstrate the method by eliciting three assistance tasks from non-experts. Wevalidate the resulting POMDP models using case-based simulations to show thatthey are reasonable for the domains. We also show a complete case study of adesigner specifying one database, including an evaluation in a real-lifeexperiment with a human actor.
展开▼
