Autominder: an intelligent cognitive orthotic system for people with memory impairment
Introduction
The world’s population is aging at a phenomenal rate. According to the United Nations Population Division, in 2000 about 606 million people, constituting approximately 10% of the world’s population, were over 60; by 2050, this percentage is expected to double to 2 billion people, or 21.4% of the population. Even more dramatic will be the increase in the percentage of people over 80, often called the “oldest old”. Today there are 69 million people in this category, constituting 1.1% of the world’s population; by 2050 the percentage will nearly quadruple to 4%, with 379 million people over the age of 80 alive [29].
It has been shown that the quality of life for people remaining in their own homes is generally better than for those who are institutionalized [23]; moreover, the cost for institutional care can be much higher than the cost of care for a patient at home. This paper describes Autominder, a system aimed at helping older adults with mild to moderate memory impairment remain in their homes longer. Many forms of memory impairment are strongly correlated with age and can make it difficult for someone to organize and regularly perform their necessary daily activities, such as taking medicine correctly, eating, drinking water, toileting, performing routine hygiene, engaging in social and family activities, keeping medical appointments, and so on.1 Autominder serves as a cognitive orthotic, providing its users (or “clients”) with reminders about their daily activities. Most existing cognitive orthotics mainly issue alarms for prescribed activities at fixed times that are specified in advance. In contrast, Autominder is capable of much more flexible, adaptive behavior. It models its client’s daily plans, tracks their execution by reasoning about the client’s observable behavior, and makes decisions about whether and when it is most appropriate to issue reminders. The current version of Autominder is deployed on a mobile robot, and is being developed as part of the Initiative on Personal Robotic Assistants for the Elderly (the Nursebot project). We are also exploring alternative platforms for Autominder, such as distributed sensors and/or wearable devices.
In the next section, we provide a description of Autominder’s architecture and its current platform. This is followed by three sections in which we discuss Autominder’s main components: the Plan Manager, the Client Modeler, and the Personal Cognitive Orthotic (its reminder generation module). We then present a brief overview of other cognitive orthotic systems, and conclude with a description of our ongoing work on the system.
Section snippets
Autominder’s architecture
To motivate Autominder’s architecture, it is useful to provide a simple example of its interaction with a client. Consider a forgetful, elderly person with urinary incontinence who is supposed to be reminded to use the toilet every three hours, and whose next reminder is scheduled for 11:00. Suppose that, using on-board sensors, the robot on which Autominder is deployed observes the person enter the bathroom at 10:40 and stay there for a period of a few minutes. Autominder may conclude that
The Plan Manager
The first of Autominder’s three main components is the Plan Manager (PM). The technology in the PM grew out of our earlier work on plan management, in particular, the Plan Management Agent (PMA), a prototype intelligent calendar tool [20]. In Autominder, as in PMA, we found that it was essential that we be able to represent a rich set of temporal constraints in the plans: for example, we may need to express that the client should take a medication within 15 minutes of waking, and then eat
The Client Modeler
The second major component of Autominder is the Client Modeler (CM). The job of the CM is to monitor the execution of the client plan, using information about observable actions of the client, as well as knowledge of what time it is and whether and when any reminders were issued. In our current implementation, the observable information is relatively impoverished—basically, the robot’s on-board sensors provide the CM only with reports of the current location of the client (what room she is
Reminder generation
The final component in Autominder is called the Personalized Cognitive Orthotic (PCO), and is responsible for deciding what reminders to issue and when [15]. In making its decisions, the PCO aims to balance four criteria: (i) ensuring that the client is aware of planned activities; (ii) achieving a high level of client and caregiver satisfaction; (iii) avoiding introducing inefficiency into the client activities and (iv) avoiding making the client overly reliant on the reminder system, which
Related systems
The idea of using computer technology to enhance the performance of cognitively disabled people dates back nearly 40 years [6], and a number of systems exist to help people with cognitive impairment perform routine activities satisfactorily. Most of these systems can be classified as either scheduling aids, which help a person manage a number of distinct activities over an extended period of time, or as instructional cueing aids, which help a person navigate the generally consecutive steps of a
Conclusion
The Autominder system as described has been fully implemented, except where noted in the text. The system is written in Java and Lisp for a Wintel platform; we also have a Web-based interface for plan initialization and update. The current version of the system has been tested in the laboratory; an earlier version was integrated on the mobile robot Pearl and included in a preliminary field test conducted at the Longwood Retirement Community in Oakmont, PA, in June 2001. The goals of that test
Acknowledgements
This work is supported primarily by the National Science Foundation Grant IIS-0085796. Some of the supporting technology was developed with funding from the Air Force Office of Scientific Research (F49620-01-1-0066). The authors wish to thank the many members of the Nursebot team for their invaluable contributions: Donald Chiarulli, Jacqueline Dunbar-Jacob, Sandra Engberg, Jennifer Goetz, Sara Kiesler, Michael Montemerlo, Joelle Pineau, Joan Rogers, Nicholas Roy and Sebastian Thrun. Earlier
Martha E. Pollack is Professor of Computer Science and Engineering at the University of Michigan. She was previously a professor at the University of Pittsburgh and, before that, a Senior Computer Scientist at the AI Center, SRI International. Pollack received a bachelor’s degree from Dartmouth College and M.S.E. and Ph.D. degrees from the University of Pennsylvania. She has conducted research and published in the areas of automated planning, plan management, agent-based systems,
References (30)
A general programming language for unified planning and control
Artificial Intelligence
(1995)- et al.
Backtracking algorithms for disjunctions of temporal constraints
Artificial Intelligence
(2000) - et al.
Planning by rewriting
Journal of Artificial Intelligence Research
(2001) - S. Carmien, MAPS: PDA scaffolding for independence for persons with cognitive impairment, 2002....
- D. Colbry, B. Peintner, M.E. Pollack, Execution monitoring with quantitative temporal Bayesian networks, in:...
- et al.
A model for reasoning about persistence and causation
Computational Intelligence
(1989) - M.M. Dowds, K. Robinson, Assistive technology for memory impairment: Palmtop computers and TBI, in: Proceedings of the...
- D.C. Englebart, A conceptual framework for the augmentation of man’s intellect, in: Vistas in Information Handling,...
Automated telephone conversation to assess health behavior and deliver behavioral interventions
Journal of Medical Systems
(1998)- et al.
Neuropage: the rehabilitation of memory dysfunction by prosthetic memory and cueing
NeuroRehabilitation
(1994)
Cited by (0)
Martha E. Pollack is Professor of Computer Science and Engineering at the University of Michigan. She was previously a professor at the University of Pittsburgh and, before that, a Senior Computer Scientist at the AI Center, SRI International. Pollack received a bachelor’s degree from Dartmouth College and M.S.E. and Ph.D. degrees from the University of Pennsylvania. She has conducted research and published in the areas of automated planning, plan management, agent-based systems, natural-language discourse, and computational models of rationality; most recently, she has been focusing on applied work on cognitive orthotic systems. Recipient of several major awards including the Computers and Thought Award (1991), and the University of Pittsburgh Chancellor’s Distinguished Research Award (2000), Pollack is also an elected fellow of American Association for Artificial Intelligence (1996), and currently serves as executive editor of the Journal of Artificial Intelligence Research.
Laura Brown completed her undergraduate education at Swarthmore College in 2000 with a degree in Engineering and a concentration in Computer Science. She then received the M.S. from the University of Michigan. She is currently attending the Ph.D. program in Biomedical Informatics, Vanderbilt University. Her research interests includes artificial intelligence, machine learning, and bioinformatics.
Dirk Colbry is working on his Ph.D. in Computer Science at Michigan State University, where his research focuses on artificial intelligence, machine vision and 3D modeling. He received the M.S.E. in Computer Science from the University of Michigan in 2002, where he worked on Autominder’s client modeler. He also received the B.S.E. in mechanical engineering from The Georgia Institute of Technology in 1997.
Colleen E. McCarthy received her Ph.D. in Computer Science from the University of Pittsburgh in 2002; her dissertation research focused on Autominder’s reminder-generation component. Her bachelor’s degree is from Kent State University. McCarthy is currently an assistant professor in the Computer Engineering and Computer Science Department at California State University at Long Beach.
Cheryl Orosz is currently pursuing M.S. and Ph.D. degrees in Computer Science and Engineering at The University of Michigan. She received the A.B. degree in linguistics from The University of Michigan in 1992. Her research interests include plan representation, temporal reasoning, plan monitoring and recognition, multi-agent systems, and probabilistic reasoning.
Bart Peintner is currently a Ph.D. candidate in Computer Science at the University of Michigan, where he received the M.S.E. in Computer Science in 2001. He has been involved in the design and development of Autominder’s client modeler, and his dissertation research focuses on advanced techniques for execution monitoring. Peintner also holds a B.S. in Electrical Engineering from Kansas State University.
Sailesh Ramakrishnan received his B.Tech. from IIT Madras and M.S. from the University of Pittsburgh. He is pursuing the Ph.D. at the University of Michigan while working at NASA Ames Research Center. His current research interests include temporal planning under uncertainty.
Ioannis Tsamardinos received his bachelor’s degree in Computer Science from the University of Crete, and his M.S. and Ph.D. degrees from the Intelligent Systems Program of the University of Pittsburgh. His Ph.D. dissertation research improved both the efficiency and expressiveness of constrained-based temporal reasoning. Tsamardinos is currently working as an assistant professor in the Department of Biomedical Informatics at Vanderbilt University, pursuing research on machine learning, feature selection, and causal induction in biomedical domains.