A POMDP Design Framework for Decision Making in Assistive Robots

This paper proposes a theoretical framework that determines the high-level cognitive functions for multipurpose assistive service robots, required to autonomously complete their tasks. It encompasses a probabilistic POMDP based decision-making strategy that provides constant situation awareness about the human and the environment by associating the robot awareness about the user with specific clusters of robotic actions. To achieve this, a method for designing POMDP models is presented herein ample to define decision making policies suitable to resolve assistive tasks through a series of robotic actions. The proposed POMDP design methodology compensates the partial and noisy sensor input acquired from the robot sensors by foreseen mitigation strategies on the robot’s decisions when a software component fails. The theoretical work presented herein is assessed over well defined robotic tasks and proved capable to operate in realistic assistive robotic scenarios.