Autominder: A Case Study of Assistive Technology for Elders with Cognitive Impairment

Abstract

Older adults said they enjoyed interacting with the robot. Autominder is an assistive technology system developed at the University of Michigan that aims to support people with cognitive impairment by providing them with flexible, adaptive, and personalized reminders about their daily activities (see Figure 1). Autominder is different from reminder systems that function in a manner similar to alarm clocks, issuing fixed reminders at pre-specified times. Through the use of artificial-intelligence technology, that is, techniques that enable a computer to reason in complex and interesting ways, Autominder constructs rich models of a person's activities, including constraints on the times and ways in which they should be performed. It monitors the execution of those activities; detects discrepancies between what a person is expected to do and what he or she actually is doing; and whether and when to issue reminders. This paper briefly describes Autominders capabilities. The following is a typical example of the way older people use Autominder. is an 80-year-old diabetic woman who has trouble with her short-term memory. She is supposed to eat a meal or snack every four hours, and she currently has an infection that requires her to take antibiotics on a full stomach. With a reminder system that emulates an alarm clock, one would have to specify the exact time at which Mrs. Jones should take her medicine. In contrast, using Autominder, Mrs. Jones or her caregiver would simply specify that she has to take her medicine within an hour of eating breakfast and, again, within an hour of eating dinner. Once Autominder recognizes that Mrs. Jones has eaten breakfast, it will know to remind her to take her medicine within the next hour, should she forget to do so. Similarly, rather than telling Autominder that Mrs. Jones has to eat at, say, 7 a.m., n a.m., 3 p.m., and 7 p.m., she or her caregiver would instead indicate to the system the upper limit of four hours between meals or snacks. If Autominder then recognizes that Mrs. Jones has eaten lunch at 11:10 a.m., it will remind her to eat again at 3:10 p.m.-or maybe a little earlier if she has indicated that she wants to watch her favorite TV show from 3:00 to 3:30 p.m. For the system to work in this way, Autominder must maintain an accurate and up-to-date model of the user's daily plan, monitor the execution of that plan, and decide about issuing reminders accordingly. Toward these ends, the Autominder system has three main components, one dedicated to each of these tasks. The Autominder Plan Manager stores the user's plan of daily activities and is responsible for updating it and identifying and resolving any potential conflicts in it. The Plan Manager is the component of the system responsible for answering the question, What is the user supposed to do? The second component, the Client Modeler, uses information about the user's actions to track the execution of the client plan. The Client Modeler thus addresses the question, What is the user doing? The third component, the Intelligent Reminder Generator, recognizes any disparities between the user's plan of daily activities and what the viser actually does, and decides whether to issue reminders. The Intelligent Reminder Generator thus answers the question, What actions should the Autominder system take to help the user be aware of the activities she needs to perform? Each of Autominder's components depends on artificial-intelligence technology. At the core of the Plan Manager is a computer program that performs what is called temporal constraint satisfaction. The program computes information about the times at which activities are to be performed: For example, when the system detects that Mrs. Jones has eaten lunch at 11:10 a.m., the temporal reasoning program propagates the constraint-or, in more familiar terms, has been programmed to remember-that Mrs. Jones is to eat every four hours and, based on the information about when Mrs. …