Supervisory control
may be defined by the analogy between a supervisor of subordinate staff in an organization
of people and the human overseer of a modern computer-mediated semiautomatic
control system. The supervisor gives human subordinates general instructions
which they in turn may translate into action. The supervisor of a
computer-controlled system does the same.
Defined strictly, supervisory
control means that one or more human operators are setting initial conditions
for, intermittently adjusting, and receiving high-level information from a
computer that itself closes a control loop in a well-defined process through
artificial sensors and effectors. For some time period the computer controls
the process automatically.
By a less strict
definition, supervisory control is used when a computer transforms human
operator commands to generate detailed control actions, or makes significant
transformations of measured data to produce integrated summary displays. In
this latter case the computer need not have the capability to commit actions
based upon new information from the environment, whereas in the first it
necessarily must. The two situations may appear similar to the human
supervisor, since the computer mediates both human outputs and human inputs,
and the supervisor is thus removed from detailed events at the low level.
FIGURE
6.1.2Direct manual control-loop analysis.
Supervisory control
system here the human operator issues commands to a human-interactive computer
capable of understanding high-level language and providing integrated summary
displays of process state information back to the operator. This computer,
typically located in a control room or cockpit or office near to the supervisor, in
turn communicates with at least one, and probably
many (hence the dotted lines), task-interactive computers,
located with the equipment they are controlling.
The task-interactive computers thus receive subgoal and conditional branching
information from the human-interactive computer. Using such information as
reference inputs, the task-interactive computers
serve to close low-level control loops between artificial sensors and
mechanical actuators;i.e., they accomplish the low-level automatic control.
The low-level task
typically operates at some physical distance from the human operator and his human-friendly
display-control computer. Therefore, the communication channels between
computers may be constrained by multiplexing, time delay, or limited bandwidth.
The task-interactive computer, of course, sends analog control signals to and
receives analog feedback signals from the controlled process, and the latter
does the same with the environment as it operates (vehicles moving relative to air,
sea, or earth, robots manipulating objects, process plants modifying products,
etc.).
Supervisory command and
feedback channels for process state information are shown in Figure 6.1.3 to
pass through the left side of the human-interactive computer. On the right side
are represented decisionaiding functions, with requests of the computer for
advice and displayed output of advice (from a database, expert system, or
simulation) to the operator. There are many new developments in computerbased decision
aids for planning, editing, monitoring, and failure detection being used as an
auxiliary part of operating dynamic systems. Reflection upon the nervous system
of higher animals reveals a similar kind of supervisory control wherein
commands are sent from the brain to local ganglia, and peripheral motor control
loops are then closed locally through receptors in the muscles, tendons, or
skin.
The brain, presumably,
does higher-level planning based on its own stored data and “mental models,” an
internalized expert system available to provide advice and permit trial
responses before commitment to actual response.
Theorizing about
supervisory control began as aircraft and spacecraft became partially
automated. It became evident that the human operator was being replaced by the
computer for direct control responsibility, and was moving to a new role of
monitor and goal-constraint setter. An added incentive was the U.S. space
program, which posed the problem of how a human operator on Earth could control
a manipulator
arm or vehicle on the moon through a 3-sec communication round-trip time delay.
The only solution which avoided instability
was to make the operator a supervisory controller communicating intermittently with a computer on the moon, which
in turn closed the control loop there. The rapid development
of microcomputers has forced a transition from manual control to supervisory
control in a variety of industrial and
military applications (Sheridan, 1992).
Let us now consider some examples
of human-machine interaction, particularly those which illustrate supervisory
control in its various forms. First, we consider three forms of vehicle
control, namely, control of modern aircraft, “intelligent” highway vehicles,
and high-speed trains, all of which have both human operators in the vehicles
as well as humans in centralized traffic-control centers. Second, we consider telerobots
for space, undersea, and medical applications.
