|
Project Summary
Adaptive automation is characterised by a reactive
rationale where the availability of automation is under
the control of the system as opposed to the human
operator. The goal of adaptive automation is to retain
the advantages of conventional automation whilst
circumventing its disadvantages. The main obstacle to
the development of adaptive automation pertains to those
diagnostic triggers underlying this system. It has been
suggested that psychophysiological variables which index
the psychological state of the operator be used as
triggers for adaptive automation. This approach is
termed ‘biocybernetics.’
The purpose of
the research proposal is to evaluate the utility of the
biocybernetic approach in a laboratory environment. The
first stage of the proposal involves the derivation of
psychophysiological algorithms as triggers for adaptive
automation. These algorithms will be incorporated into
a biocybernetic testbed linked to a simulation of
flightdeck activities. Once this testbed has been
established, a human factors study is planned to
investigate criterion for biocybernetic adaptation, the
usefulness of biocybernetic adaptation for an automated
system. This is a basic research programme into an area
of novel technology which may have important
implications for the transportation industry.
Background
System automation
represents an important development for ‘intelligent’
technology. The function of system automation is to
computerise tasks or task elements, which were
previously under the direct control of the operator. The
concept of system automation is exemplified by the
‘autopilot’ facility available on the flightdeck. It
was anticipated that system automation would improve the
aviation industry via several possible means (Mouloua,
Deaton, & Hitt, 2000):
-
to compensate for the
psychological limitations of the pilot by allowing
fast and accurate control of multiple, simultaneous
tasks
-
to alleviate the psychological
consequences of high mental workload and sustained
performance such as stress and fatigue
-
to enable more precise flight
planning and fuel utilisation
-
to reduce
operational costs
-
to improve
safety.
There are several
possibilities for the control of system automation. The
human operator may control the availability of
automation. For example, a pilot may engage and
disengage the ‘autopilot’ at his discretion. One
alternative is to make automation available on a pre-set
schedule, e.g. ‘autopilot’ unavailable during the
landing phase of the flight or ‘autopilot’ facility is
available for a maximum period of continuous use. The
development of adaptive automation represents an
extension to both conventional approaches. Under
conditions of adaptive automation, the availability of
automation is dynamic, based on real-time variables and
under system control (Parasuraman, Sheridan, & Wickens,
2000). The goal of adaptive automation is to retain the
advantages of system automation whilst circumventing its
disadvantages. An adaptive system controller would make
automation available in response to real-times measures
of operator stress or performance degradation. This
reactive rationale has the advantage of tailoring the
availability of automation to the individual operator
and the specific task situation.
The key obstacle
to the implementation of adaptive automation is the
development of those monitoring/diagnosis algorithms
that underlie the system. The capability of this
monitoring system sets practical limits on the
possibilities of adaptive automation. The most
fundamental research question concerns the
identification of variables to function as triggers for
adaptive automation. Several types of real-time data
are available as potential triggers for the adaptive
controller, i.e. the component that controls the
availability of automation. An adaptive controller
could monitor the quality of pilot performance and
automate if errors or long response times were
detected. Alternatively, the controller may monitor the
external demands of the flight as indexed by
environmental variables such as wind speed, visibility
etc. It has also been suggested that
psychophysiological variables, which index the internal
state of the operator, be used as real-time triggers for
the adaptive controller (Byrne & Parasuraman, 1996).
Changes in
operator alertness as indexed by cortical activation
from the Electro-Encephalo Graph (EEG) or fluctuations
in operator stress quantified by heart rate offer
several advantages as potential triggers for automation.
Psychophysiological changes are continuous in real-time,
whereas performance-based triggers may be discrete and
sporadic. The psychophysiological status of the
operator may also be monitored in the absence of any
overt inputs or behaviour. This is particularly
important when the operator works with a completely
automated system and task engagement is totally covert.
Finally, psychophysiological variables are extremely
sensitive and may even predict performance impairment,
if collected and analysed appropriately.
The use of
psychophysiological variables as triggers for adaptive
automation has been termed ‘biocybernetics’ (Prinzel,
Scerbo, Freeman, & Mikulka, 1995). This proposal is
concerned with the development and evaluation of a
biocybernetic system in a laboratory, using a simulation
of flightdeck activities.
Current Situation
The development
of a biocybernetic system rests on the assumption that
certain psychological states such as fatigue and stress
have a tendency to impair human performance. There is a
wealth of empirical evidence to support this view, see
Hockey (1986) for review. A second assumption holds
that stress and fatigue can be accurately measured in
real-time by psychophysiological variables. This view
is also supported by reviews of applied
psychophysiological research, e.g. Kramer, Trejo, &
Humphrey (1996).
Both assumptions
formed the basis of original research conducted by Pope,
Bogart, & Bartolome (1995). These authors used the EEG
as a psychophysiological index of task engagement. This
index was measured in real-time and used to adaptively
automate a tracking task. Pope and his colleagues
experimented with several global EEG variables to
measure task engagement and act as triggers for adaptive
automation. They reported a ratio formula of EEG
bandwidths that yielded the most sensitive index of task
engagement.
Pope et al (1995)
also compared positive and negative biocybernetic
feedback systems. In the former case, the tracking task
was automated if the adaptive controller detected
reduced task engagement. The negative feedback system
worked in the opposite direction, automating the
tracking task when task engagement was deemed extremely
high (indicating task-induced stress) and turning the
task over to manual control when task engagement was
low. Pope et al (1995) reported that levels of task
engagement (as indexed by EEG) were unstable under
positive feedback conditions. This finding was
anticipated given that positive feedback tends to
inflate the underlying trend of task engagement, whereas
negative feedback is designed to counteract this trend.
The line of
research initiated by Pope et al was continued in a
series of experiments conducted by Freeman et al
(1999). These authors replicated and extended the main
findings of Pope et al (1995), indicating that adaptive
automation was capable of improving the quality of
tracking performance. This work also studied the
quantification of criterion to trigger adaptive
automation. The original Pope et al studies indexed
positive or negative changes in EEG task engagement
using a slope criterion, i.e. a positive change from one
epoch to the next indicated increased engagement and
vice versa. Freeman et al (1999) contrasted this
criterion with an absolute criterion, i.e. a positive or
negative change relative to a resting baseline. Their
data indicated that adaptive automation enhanced
tracking performance to a greater extent when an
absolute criterion was in operation. However,
subsequent work from the same laboratory produced data
that conflicted with this hypothesis (Freeman, Mikulka,
Scerbo, Prinzel, & Clouatre, 2000). These issues of
criteria development remain to be resolved via further
study.
Research on
biocybernetics and tracking performance was extended by
Prinzel et al (2000) who used a multiple task
environment as an experimental test-bed. The
Multi-Attribute Test (MAT) Battery was developed by NASA
to emulate flightdeck control activities. The MAT
Battery is composed of three sub-tasks to simulate the
types of monitoring, manual control and resource
management tasks performed by crewmembers on a
flightdeck (Comstock & Arnegard, 1992). Prinzel et al
(2000) used the EEG index of task engagement coupled to
adaptive automation of the manual control task under two
experimental manipulations: (a) single control task vs.
multi-tasking, and (b) adaptive automation vs. static
automation, i.e. the former based on real-time EEG
measurement whereas automation was activated at pre-set
points in the latter case. They reported that
biocybernetic adaptation was responsive to increased
mental workload due to multiple task performance. In
addition, both performance errors and subjective
workload were significantly reduced when automation was
based on biocybernetic control. This study provides
clear evidence of the potential benefits associated with
biocybernetic control of automation.
The small number
of studies described in this section is indicative of
the relative paucity of available data on this topic.
Several important areas remain to be investigated,
including:
-
The
development of valid, psychophysiological algorithms
capable of predicting performance errors and
subjective fatigue or stress.
-
The
incorporation of other psychophysiological
indicators such as ECG, respiration rate, blood
pressure as possible triggers for adaptive control.
-
The
suitability of slope versus absolute criteria for
biocybernetic control
|
Project Title
"Biocybernetic
Control of Adaptive Automation"
Sponsors
EPSRC
Partners
Centre
for Applied Psychology
Liverpool John Moores University
James Parsons Building
Byrom Street
Liverpool
L3 3AF
School of Computing and Mathematical Sciences
Liverpool John Moores University
James Parsons Building
Byrom Street
Liverpool
L3 3AF.
Tel (+44) 1524 593801
|
