Intelligent Monitoring in a Robotic Assistant for the Elderly

Abstract

The NurseBot project is developing a mobile robot that is intended to assist elderly people suffering from mild cognitive disorders in their everyday life (see http://www.cs.cmu.edu/~nursebot). One of the main components of the project is intelligent reminding, which is useful when an elderly person has mild memory problems. The robot possesses information about the elder’s daily activities, and monitors their performance, providing reminders when needed. The reminders may address activities that are critical to the elder’s health and safety (taking their medications), as well as activities that support the elder’s general happiness (their favorite TV program). An intellig ent reminder system must be able to analyze its user’s plans and compute a monitoring plan of its own. One of the main aspects of intelligent monitoring is to identify which activities should be monitored. It is infeasible to monitor everything, for several reasons. First, the robot may not be able, or might choose not to follow its user around all the time, for example, it might not enter the user’s bathroom. Second, the kinds of and the accuracy of the sensors may limit the ability of the robot to m onitor certain activities. Third, too much monitoring may annoy the elderly user, who may be attentive to and punctual about certain activities, only requiring monitoring for other activities. Prior work on selecting activities to monitor during plan exe cution (e.g. Pollack and McCarthy 1999) has not focused on these aspects of the problem, but instead has distinguished between environmental changes that influence an existing plan in some way, and those that do not. The intelligent monitoring system we a re developing learns which activities need to be monitored, and when reminders need to be issued. It is initially provided with a detailed representation of the user’s plans, which may be modified, via additions, deletions, or changes, as time