Abstract
Multi-parameter
closed-loop controllers for the upper limb were operated by a C4-level SCI
patient for writing. Two versions of the controller: (i) PID and
(ii) adaptive, enabled writing with a
pen, although neither system was easy to us
1.
INTRODUCTION
Two Beer Sheva upper limb neuroprostheses, both of
them non-invasive, were developed during
the 1980's in open-loop: for C4 lesion-level SCI1 and for C5 lesion-level, stroke, CP, MS, and TBI2. The latter
C5 system evolved into, and is now marketed in Europe and Asia as the
Handmaster, and in
The
C4 neuroprosthesis restored a variety of ADL.
Writing was difficult however in open-loop, requiring continuous
adjustment (by system-user voice command) of low-level functions required for
pen manipulation: gripping and pen-paper pressur
Both
classical control and "rule-based artificial reflexes", pioneered by
Tomovic3 were implemented in the
controller. It transpires that both the
author and Brian Andrews were inspired to implement Tomovic's theories (on the
upper limb and lower limb respectively) after receiving impromptu
"beach" seminars at the Portoroz conference in 1983.
The
controllers were designed to take responsibility for the system user and his
interaction with the environment, keeping the system under control even under
adverse conditions such as extension spasms of the limb; and allowing standby periods and
"failsafe" response to operational breakdowns. These are requirements
for unsupervised neuroprosthesis use in the hom
Since
this period no further work has been carried out on the Beer Sheva C4
neuroprosthesis.
2.
METHODS
2.1 System Hardware
Hand prehension/release, wrist
flexion and elbow extension are generated by 12 surface electrode arrays
positioned on the upper arm, forearm and hand intrinsics, each comprising 22
electrodes covering a target muscl
Wrist
joint extension is elicited by a mechanical spring. The limb is supported in an arm support
attached to the wheelchair which allows a limited degree of hand movement in
the horizontal plane from voluntary control of shoulder girdle muscles. Voice input commands provide a logic-based
command system.
Fig. 1: The Beer Sheva C4
Neuroprosthesis
A closed-loop control capability (figure1), was
added to the system to enhance the ability to manipulate a pen and writ
2.2 Software Controllers
2.2.1 General
Control
The system user inputs
high level commands such as "GRIP", "DOWN" or "UP". Implementation of the commands
and control of the hand/pen/paper interaction is carried out by the
controller. Several logic-based
fault-sensing and correcting algorithms were added to the controller to answer
specific needs and events that can occur in the home environment. For example should the pen slip out of the
hand, the current intensity would rise unabated in order to attempt to increase
the (non-existent) grip strength. To
identify and answer this occurrence, a characteristic combination of system
conditions are sensed, the stimulation current intensity is zeroed, and
the system is put on "standby" until further commands are input.
Sensor input patterns detecting extension spasm result in a temporary reduction
of the stimulation intensity until the spasm has passed, at which time the user
can carry on writing. Algorithms were
also included to account for sensor nonlinearities and drift.
Hand prehension: opening/closing motion is controlled in
open loop. On command the stimulation
intensity to the finger and thumb extensors is incremented to open the hand. A global stimulation intensity constant can
be incremented or decremented by the system user. A further input command closes the hand by
decrementing the stimulation intensity to the extensors and incrementing to the
flexors with some intermediate coactivation.
A command "DOWN" to lower the pen to the paper now switches to
the closed loop regim
Active
control is shared between the two degrees of freedom according to their
requirements. Pen/paper force is far more difficult to
control, and requires a relatively high sampling rate sophisticated
strategy. Gripping force is on the other
hand is more tolerant and simpler to control, and a regime was developed whose
aim is to keep the gripping force within reasonable limits, while responding
fast to perturbations in the pen/paper forc
2.2.2 Gripping force
A "target band" is predefined within
which the force is allowed to drift. If
the gripping force strays outside the target band, the stimulation current
intensity to the muscles involved is increased or reduced by fixed increments
until the force re-enters the band.
Environmental
interactions during writing cause the gripping force and the distribution of
this force between the fingers and thumb to be perturbed by changes in the
pen/paper force vector, particularly when the pen is lowered on to the paper,
or during "vigorous" writing.
These perturbations are transient and are filtered out.
2.2.3 Pen-paper force: PID Control
The primary goal of the
controller is here to hold the pen to the paper continuously, with minimum
contact force while the pen is moved randomly in a horizontal plane during
writing. The horizontal pen movement
generates a large stochastic perturbation on the monitored
system force output. The force
perturbation is transient in nature, but can reach amplitudes many times that
of the target forc
where DI is the
stimulation current increment, e is the error between the measured and the target pen/paper force, and KP,
KD, and KI are constants. Iterative
hand-tuning of the equation constants was carried out experimentally
2.2.4 Pen-paper force: Adaptive Control
The PID equation constants are adapted
as a function of the variance of the disturbance in the pen/paper force, thus
assessing the mood of the person and to a lesser extent the operating
characteristics of the system plant. The
pen/paper force trace during continuous writing at typical speed generally
takes the form shown in figure 2.
The pen/paper force feedback is contaminated by high amplitude
stochastic disturbances as the pen is moved horizontally in different
directions and speeds during writing. It
is difficult to feedback on-line information on the success or failure of the
control system, as this imposed stochastic force drowns reliable information on
the response characteristics of the system.
Rather the pen/paper force feedback primarily indicates the
characteristics of the volitional writing motion, and to a lesser extent, the
ability of the controller to cope with it.
The character of the pen/paper force is assessed periodically at
approximately 10 second intervals by the average of the wave amplitude and of
its duration. Adaptations are carried
out on the equation parameters according the conditions listed in table 1. Limits were placed on the range of values of
the PID equation parameters. These
limits covered realistic range of values for each parameter. While the pen is
raised between strokes, the assessment and adaptation procedure is frozen.
The controller
effectively adapts to the mood of the system user, the type of writing, the
type of pen and paper, and to its own performanc
3. RESULTS
3.1 PID Controller: Figure 3 shows the gripping force and pen-paper force and also the
stimulation current intensities as the pen is lowered to the paper and writing
commences.
.
Fig 3: PID Controller Performance during Writing
Poor
controller performance manifests itself in several possible ways: as a
sluggishness of response, resulting in slow correction of pen/paper contact
forces, "digging in" of the pen into the paper or wedging of the pen
into the paper during the forward stroke,
and the tendency of the pen to "hover" slightly above the
paper for short periods, particularly on the return strok
These phenomena can be eliminated by tuning the control
equation constants KP, KD, and KI to suit the prevailing system conditions.
3.2
Adaptive Controller:
Figure 4 shows an example of writing with the adaptive
controller.
Fig. 4: Writing with the Adaptive Controller
The letters are on average 1.5 cm in height, and the note
took about 10 minutes to writ
The
conditions for adaptation of the controller parameters in effect caused an increase in the
"springiness" (increased gain with reduced damping) of the control
where the system user writes slowly and carefully, while "deadening"
(reduced gain with increased damping) the control where the system user writes
fast and jerkily. These adaptations
tend to shift the system poles away from the prevailing wave frequency of the
writing to avoid system resonance, correcting the pen/paper force during slow
careful writing and drawing, but effectively filtering out the force waves
during fast writing and scribbling.
4. DISCUSSION AND CONCLUSIONS
The PID controller was found to behave fairy
robustly, but to occasionally experience an "off day" where the
preset constants controlled the system unsatisfactorily.
Assessment of the adaptive
controller on one patient shows the controller to respond as intended,
"deadening" the system in response to a lively operator, and
"spicing it up" where the operator writes lethargically.
References
.[1] Nathan, R.H., Ohry, A., Upper limb functions regained in the C4 quadriplegic by a computerized neuromuscular stimulation system, Arch. Phys. Med. & Rehab., 71: 6, pp. 415-421, 1990.
[2]
Nathan, R.H., A non-invasive FES system for restoration of hand function in C5
quadriplegia and CVA, Proc 2nd Int FES Symp, pp 128-133, Sendai, Japan,
1995.
[3] Tomovic, R., Systems approach to the control of artificial limbs, in
The control of upper extremity prostheses and orthoses, ed. Herberts P, Kadefors
R, Magnusson R, Peterson I, 254-257, Thomas, Springfield, 1974.
[3] Nathan, R.H., and
Gabai, Y., FNS of the paralysed upper limb: closed loop control for writing,
in Electromyographical Kinesiology, eds.
[5]
Nathan, R.H., Control strategies in