Compensatory Tracking

[From Bruce Abbott (950327.1750 EST)]

O.K., you've had your philosophical interlude; let's get back to science.
In an effort to understand SDTEST3 performance, I've created a simpler
program called CTRACK1, which looks almost identical to SDTEST3 except that
there is only one target (at screen center) instead of two, and the cursor
stays green throughout the run. The idea is to gather data on the
compensatory tracking part of SDTEST3 sans SD disturbance for the purpose of
building a control model for this part of the task and determining its
parameters, under the assumption that they will be the same as in SDTEST3's
lower-level control system.

Perhaps a more complex model is not needed at this stage, but I've been
wondering whether we need to be taking account of such things as the
effective mass of the arm-mouse system and the frequency spectrum of the
disturbance. To investigate these aspects of performance, I've used a
simple sine-wave disturbance and varied the period of the wave. Thus far
I've tried the following series: 75, 150, 300, 600, and 1200 "ticks"
(1/60th sec; 600 corresponds to a period of 10 sec). The maximum
disturbance was scaled to plus or minus 60 and the mouse sensitivity was set
to 0.5 (i.e., mousex div 2). What follows are summary stats for 5 runs at
each period:

Period k sd k hi k/lo k rms sd rms
  1200 0.13 0.04 1.6 3.5 0.4
   600 0.12 0.02 1.4 3.9 0.3
   300 0.12 0.01 1.6 6.9 0.8
   150 0.16 0.02 1.3 12.4 1.0
    75 0.16 0.08 3.9 33.4 7.2

At the longer periods, the main problem with tracking accuracy is a certain
degree of slip-stick friction in my arm joint and a bit of muscle
twitchiness, especially when the arm is moving away from the body. As the
period shortens, I find myself tending to lag behind the disturbance and
overshoot the peaks of the sine wave. At the shortest period (75/60, or
1.25 sec), I had serious difficulty keeping in phase with the disturbance; I
more-or less gave up trying to actually track, opting instead to try
generating a movement that was more-or less sine-like and at the same
frequency and amplitude as the disturbance.

The most systematic effect of period was on the rms error, which was
inversely related to period. The variability of performance also increased
with decreasing period. There is some suggestion that k increased at the
shortest periods, but this should not be overemphasized; within each set of
5 runs at a give period the k-values tended to vary over a fair range (see
hi k/low k ratios).

I'm wondering if performance under the sine-wave disturbance might be
analogous to a "forced oscillation" situation. Picture a spring whose top
end is being moved vertically in a sine-wave motion across a given distance.
Attached to the bottom end of the spring is a mass; the spring and mass
constitute an oscillating system with its own natural frequency, one which
may differ from the driving frequency. As the upper end of the spring is
driven upward, the mass on the other end resists being accelerated (a =
f/m), so the mass lags behind the driving movement. Similarly, when the
spring is being driven downward after reaching its upper peak, the mass on
the other end may still be moving upward. The actual wave traced by the
movement of the mass would depend on the mass, the stiffness of the spring,
the natural frequency of the system, and the driving frequency. At high
enough frequencies the mass would hardly move at all, whereas at low enough
frequencies it would very nearly track the movement of the upper end of the
spring.

Given that I'm neither a physicist nor an electronics engineer, I'm not
really in a good position to figure out how to apply such a model to the
tracking situation, or even better, to determine whether the analogy I'm
proposing is close enough to be worth the attempt. It seems to me that I'm
essentially talking about frequency response analysis. Such an analysis
might provide a way to predict how the participant would respond to the
random disturbances we normally use in these tasks. What say you engineers?

I'm also planning to vary other parameters, including the mouse sensitivity
and maximum excursion of the disturbance, to determine what, if any, effects
these have.

Regards,

Bruce