paper presents preliminary work on the design of control
systems for pedestrian mobility aids for the elderly. The elderly
are often restricted in their mobility and must rely on canes,
walkers and wheelchairs for locomotion. Restrictions in mobility
lead to a loss of independence and autonomy, as well as a
decrease in muscular strength.
This paper focuses on design of intelligent wheeled walkers. By
allowing the user varying degrees of control, from complete to
collaborative, these walkers afford the user with the feeling of
control, while helping to increase the ease and safety of their
daily travels. The control systems of these walkers differ from
those of other mobility aids and mobile robots because they must
both assist in mobility and provide balance and support. These
functions must be performed in a tight loop with a human whose
input may be difficult to predict.

Introduction


The worlds elderly population is increasing dramatically.
In the US, there are more than 34.8 million seniors over the
age of 65. Furthermore, in only 30 years, this number will
more than double to 70 million [10]. In Japan, already the
nation with the highest percentage of seniors on earth, it is
estimated that 1 in 5 people will be seniors within 10 years
[17]. At the same time the costs of health care, including
caring for the elderly, could rise from its current $1.3
trillion to over $4.0 trillion [6]. If robotic technology can
be used to enable the elderly to remain independent,
significant costs could be saved and the quality of life
would be improved for these people. The attainable cost
savings are significant: for every single month that we
delay the transition of the elderly population into nursing
homes, the US economy saves over $2 billion [1].
Many seniors have mobility impairments that cause a
downward trend in their quality of life. Lack of
independence and exercise can have dramatic results.
Walkers are used more than any mobility aid except the
cane [15]. This paper describes our work in designing
control systems for wheeled walkers (rollators) to aid in
pedestrian mobility of the elderly. We describe various
shared control strategies that enable an elderly user and
their walker to collaborate on movement, increasing the
ease and safety of the users travels.

Copyright © 2000, American Association for Artificial Intelligence
(www.aaai.org). All rights reserved.


Design Issues
The control system of our pedestrian mobility platform is
distinct from the control systems typically used in mobile
robotics or other assistive devices because it must provide
both mobility and balance/support. The difficulty in
designing our walkers control system is that the
collaborative aspect of moving the user must be taken into
account.
This collaboration aspect can be viewed as the fusion of
two distinct control systems via the physical frame of the
walker. One system is the human, providing the motive
force and steering the walker towards their goal. The
second control system is the steering and braking provided
by the walker to avoid obstacles and prevent falls.
These two systems are not independent of each other and
care must be taken to insure that errors in the systems do
not compound. A walker user anticipates both the
movement of the walker frame and the movement of their
own body with each step. If the walkers movement differs
significantly from expectation, a fall can result. In addition,
if the users expected body motion does not match that
produced by their muscles, a fall can occur if the walker is
not in a position to allow the user to catch their fall. This
means that the walker must strike a balance between
placing itself in a position anticipated by the user and
placing itself in a position that actually supports their
musculo-skeletal system.
The control system must take into account more than just
the users balance. Users expect that the walker will
maneuver in the way that they push it. The control system
must not make the user feel as if the walker is unresponsive
or non-obedient. The walker must also help the user reach
their destination. Providing the feelings of independence
and control to an elderly user, who may have decreased
sensor acuity, decreased reaction times, and increased
muscle spasticity, is a primary goal of this work. Our
success can be measured by the degree to which the control
system can help the user do what they mean to do, as
opposed to what their physical input might suggest.
A critical design issue is the ability to integrate the two
control systems to provide a smooth sense of shared
control. If the autonomous control system in the walker
makes abrupt changes to the speed or direction of travel of
the walker, it could easily cause the precise problems it is
designed to solve. In addition, the operator cannot be
expected to enter complex path data into the system while trying to navigate. Rather the walker must attempt to use
small changes in the operators choice of direction and
speed to maintain an estimate of the operators goals. Using
these estimates, the walker control system must adapt its
control in a smooth, collaborative manner.
A Variety of Control System Designs
The Medical automation Research Center (MARC) is
designing several walker control systems with different
capabilities. As mentioned previously, all of our walkers
lack propulsion and rely on their user to provide forward
motion. All control systems can actuate the walkers brakes
and most have the ability to steer, by controlling the
orientation of walkers front wheel. The walkers are
equipped with sonar, infrared sensors, and wheel encoders,
to determine their location in the local environment. The
control systems help walker users avoid obstacles and
cliffs. Cliffs are sudden drops in the users walking
surface, such as stairs or a sidewalk curb, which can cause
the user to fall. In addition, some control systems can
estimate their global position within a known environment
(like a private home, rest home, or health care facility) to
help guide the user to particular locations. The design of a
control system that meets the above criteria depends on the
service(s) that the walker is to provide. We are currently
developing four types of control systems for the walkers
(see figure 1).

Figure 1. MARC Smart Walker
These control systems begin with informational support
only, and then add layers of control. These control layers
progress from safety control, through simple navigational
aid, to more complex goal achievement assistance. Each
successive layer builds additional competencies into the
walker in a subsumptive model based on that proposed by
Brooks [4]. In effect, these layers of control act as a set of
possible goals for the walker, and based on the evolving
user inputs and sensor data, the walker reprioritizes these
goals [9] to collaborate with the user.
Warning System Only
The warning system only control system cannot steer or
brake the walker; it can only alert the user that there is
impending danger. Small motors mounted in the walkers
handles cause them to vibrate when the sensors detect a
nearby obstacle. We feel that this has an advantage over an
audible warning because it is felt only by the user and does
not call attention to them. However, a disadvantage is that
the user must scan the nearby area to locate the obstacle
detected by the sensors.
This walker has no means of effecting its own motion,
and therefore collaborates with the user in only the simplest
manner. It will be used as one of the baselines in our user
tests.
Safety Braking Only
This type of control system can brake the walkers wheels,
but cannot steer. The walker frames original, bicycle-style,
hand brakes have been augmented with a lead-screw system
that can depress the brake pads. This system can be
triggered whenever the walker comes to a stop, or when an
imminent collision is detected. The brake system allows the
walker to be slowed gradually, as an obstacle is
approached, stopping at some minimum distance.
The control system has the goals of preventing collisions
and keeping the user upright. Preliminary experiments with
a solenoid braking system showed that fast-firing brakes
can be disconcerting to the user. Instead, brakes should
come on and off slowly so that the user feels the resistance
increase in stages. This leads to these control rules:
If the walker is in motion (as
determined by the wheel encoder),
the force applied to the brake pads
is inversely proportional to the
distance to obstacles.
(1)
If the walker is stopped, the brakes
should be fully applied to provide
a stable base on which the user can
rest
(2)
When the walker is stopped and the
user wishes to move again, the
brakes should come off slowly to
prevent the walker from lurching
forward
(3)
Rule 1 is self-explanatory. Rule 2 is used to provide a
stable frame configuration when the user stops to rest. Rule
3 raises two interesting questions. First, how to distinguish
between the forces applied to the walker while leaning
against it to rest and forces that indicate that the user
wishes to move and second how to deal with users with a
slow or halting gait. The first question is difficult and the
subject of our work in designing force sensors