Professional Documents
Culture Documents
The strategies used by operators to control machines have long been of keen interest to
the flight test community. Research into these strategies has, until recently, assumed that
pilots control their aircraft solely to maintain some condition. In 2004, boundary avoid-
ance tracking (BAT) was recognized as a strategy wherein pilots control their machines
to avoid a condition. While BAT has provided some insight into pilot-induced oscillations
(especially those involving hazardous boundaries) it has more recently been utilized at the
USAF Test Pilot School (TPS) to evaluate aircraft handling qualities using a subjective
build-up approach. An abbreviated history of BAT research at the USAF TPS is presented,
from initial computer modeling to in-flight testing on the variable stability NF-16D VISTA
aircraft. Earlier tests have demonstrated the existence of boundary avoidance tracking,
produced results that suggest modeling of pilot boundary avoidance response is possible,
and illustrated how tracking data gathered using boundaries may aid in characterizing air-
craft handling qualities. Furthermore, the VISTA test showed a good correlation between
the Cooper-Harper task results and a BAT flight test technique (FTT), strongly suggesting
that an FTT utilizing boundaries may be sufficient to objectively characterize an aircraft’s
pilot-in-the-loop capabilities for some tasks. This success and others have led to the incor-
poration of BAT flight test techniques into the USAF TPS handling qualities syllabus to
educate students about the effects of pilot aggressiveness and to provide an additional tool
for their future operator-in-the-loop tests.
Nomenclature
xb Displacement from a boundary
tb Instantaneous time to a boundary considering only displacement and rate
Kbm Maximum value of boundary feedback
tmin Minimum time-to-boundary for null feedback
tmax Maximum time-to-boundary for maximum boundary feedback (Kbm )
τb Pure time delay before boundary feedback is applied
1 of 19
This material is declared a work of the U.S. Government and American Institute
is not subject of protection
to copyright Aeronauticsin theand Astronautics
United States.
an algorithm.2 Their algorithm is typical in that it is designed to optimize machine control, not duplicate
or explain a pilot’s control actions.
In 2004 the author published a paper challenging the idea that pilots only attempt to maintain conditions
and proposing that pilots also engage in a type of tracking he called “Boundary-Avoidance Tracking.”3 In
this type of tracking, the pilot controls the aircraft in relation to an undesired aircraft state. Perhaps the
best illustration for this is the bicycle/beam analogy.
The analogy of riding a bicycle across a beam has been a popular way to teach the concept of “gain” by
graphically illustrating how the difficulty of the task changes the rider’s inputs. Imagine the task of riding a
bicycle down a straight line painted in the middle of a parking lot; this illustrates the low-gain rider. If you
take that path and elevate it to a deadly height (i.e. a bridge without rails) then make the path narrower
and narrower, the rider will clearly increase his effort to stay on the painted line; an increase in gain. If the
path is narrow enough, the rider’s gain becomes so high that he can no longer remain stable and oscillates
off the narrow beam.
Traditional handling qualities analysis assumes that the risk of falling off the narrow beam increases the
rider’s gain, perhaps to the point of triggering a pilot-induced oscillation (PIO). BAT theory states that
the rider actually controls in relation to the edges. As the rider approaches an edge he is said to engage in
boundary avoidance tracking, where the boundary to be avoided is the edge of the beam. A singular instance
of BAT at one edge can drive the rider into the other edge where the process repeats until the bicycle and
rider tumble into the abyss.
2 of 19
tmin − tb
Kbm (1)
tmin − tmax
tmax < tb < tmin Pilot response increases lin-
early with decreasing tb where
tb = xb (dxb /dt)−1 (2)
I.B. The Path to the Current Understanding of BAT Theory and Application
The original computer and desktop simulator research into boundary avoidance tracking was conducted by
the author under the auspices of the USAF Test Pilot School. BAT has been the subject of several student
flight test projects at the USAF TPS and their associated master’s degree thesis work with the US Air
Force Institute of Technology (AFIT). The author has also been informed of some applications of BAT flight
test techniques in US Air Force and Navy developmental test programs, but these will not be addressed in
this paper. New test techniques were developed and fine-tuned over the course of the USAF TPS and AFIT
projects, and data from the projects encouraged the examination of new methods for quantifying pilot inputs
during a tracking task.
The development of hypotheses, conduct of research, and fine-tuning of test techniques were necessarily
interlaced and, aside from the research projects, were not neatly parsed in time. They will be presented in
the general order in which they inspired and supported each other. The current understanding of boundary
avoidance tracking and its use in handling qualities evaluation is the result of these several year’s work and
has now been successfully woven into the USAF TPS curriculum, providing both a theoretical basis and a
framework for clear and repeatable handling qualities demonstrations.
3 of 19
(c) Both boundaries create a single input. (d) Stable oscillation between boundaries.
1. Unstable boundary escape oscillations tended to grow ‘explosively’ until reaching the bound-
ary tracker gain limits, Kbm .
2. Feedback inputs for a boundary escape oscillation that has diverged to the gain limits are
characterized by stop-to-stop inputs.
3. Boundary escape tracking produces extremely nonlinear (‘clifflike’) results. Very tiny vari-
ations in gain, time delay, or boundary awareness parameters in the boundary-tracking
feedback loop marked the transition from a moderately damped boundary escape response
to rapidly divergent oscillations.
4. Increased feedback delay was an especially powerful driver of boundary escape oscillations.
5. Unstable point-tracking oscillations can rapidly transition to catastrophic boundary avoid-
ance oscillations once boundary awareness is achieved. The transition may be marked by
an explosive increase in feedback (inceptor) inputs.
6. Boundary escape PIO can occur where point-tracking PIO is not present. If the boundaries
are sufficiently tight and/or the increase in gain brought by boundary escape tracking is
sufficiently greater than the normal gain for the point-tracking task, a boundary escape
PIO can quickly arise from a disturbance large enough to assault one of the boundaries.3
Perhaps the most important result from this work was the response of the test pilot community following
the author’s presentation of the results at the 2004 Symposium of the Society of Experimental Test Pilots.
Numerous test pilots made a point of telling the author their PIO stories and how boundary avoidance
tracking successfully described their focus during these events. Many other pilots provided descriptions
of different types of boundaries and how these boundaries affect their flying. One particularly intriguing
4 of 19
5 of 19
6 of 19
7 of 19
V. A Limited Investigation of
Boundary Avoidance Tracking (HAVE BAT) 10
“HAVE BAT” was a student test management project conducted at the USAF Test Pilot School in the
Spring of 2006 to examine boundary avoidance tracking in flight. A T-38C modified with a data acquisition
system, including several video cameras, was used as the test aircraft. A target platform was created by
marking a T-38A with a set of visually clear boundaries for pitch tracking in close formation flying. Figure 6
shows the target aircraft. Pilots were tasked with maintaining a vertical position on the target aircraft by
keeping the wingtip between markings on the fuselage of the target. Four boundaries were available; the two
marked with unbroken and hashed lines, the upper and lower parts of the intake structure, and the center
of the USAF insignia (for tracking with null boundary displacement).
The objective of the HAVE BAT test was to duplicate the earlier simulator study in flight by giving pilots
a formation z-axis displacement tracking task. Numerous practical difficulties made that goal unachievable in
the time available. The test team intended to use video of the lead aircraft to determine the tracking pilot’s
view of the boundaries but the video quality could not support automated processing, leaving insufficient
time to analyze most of the data runs. For those data runs that were analyzed (by manually extracting
data from the video frame-by-frame), comparison of the pilot’s stick inputs to BAT theory was not possible
because the ideal position of the stick at each moment in time was unknown. Without knowing the point
from which to apply the boundary feedback, the BAT parameters could not be determined. The test was not
without significant successes, though; most notably that the student test pilots could condition themselves
to treat the artificial boundaries as if they were real and that the result on their task was consistent with
BAT theory. In all cases, as the boundaries were made more restrictive, each pilot would reach a point where
they would PIO. The pilots also noted that there was a clear cognitive transition between point tracking and
boundary avoidance tracking—finding that transition in the inceptor data is a different problem altogether.
Although somewhat disappointing in terms of the original objectives of the test, the HAVE BAT program
8 of 19
9 of 19
10 of 19
Figure 10. HUD Displays for the BAT DART tracking tasks (taken from VISTA HUD video).
In the LAMARS and VISTA, each subject conducted 8 total test runs. Four different models were
flown using both a Cooper-Harper task and a boundary avoidance task. These models included two Level
I (acceptable) models, a Level II (tolerable) model, and a Level III (unacceptable but controllable) model.
With the exception of the pitch response, all tasks were identical flight path angle tracking tasks. A sum-
of-sine pitch input to an F-16 model was used to create a flight path that the subject tracked during the
task. Figure 10 illustrates the CH task and BAT task displays as seen by the subject in the heads-up display
(HUD). For the workload buildup task, the task started with wide boundaries displayed in the HUD and the
pitch tracking task would repeat every minute, with the boundaries closing-in by 25% every step. (Figure 10b
depicts the display at one of the intermediate boundary settings.) The task continued until either boundary
11 of 19
Figure 11. Comparison of the same tracking task (the sixth step in the workload buildup) between the Level II and
Level Ia model.
12 of 19
13 of 19
Task Performance Establishes whether or not a pilot Establishes the best performance the pi-
can accomplish a particular task. lot was capable of in a particular task
Learning Curve Learning curve is observed through Learning curve is observed through im-
Assessment improved ratings as the task is re- proved performance as the task is re-
peated. peated.
able to accomplish the full 60 second duration of the sub-task. (The final step from each run was excluded
because runs were typically terminated by a boundary excursion at the first significant challenge. This
resulted in a computed RMS error applicable only for the shortened task and not directly comparable to the
results from completed 60 seconds workload buildup steps.)
Figure 13 shows how performance improved as the boundaries were brought together. Note that the
subjects allowed more error in flight than in the simulator–this is probably due to the consequences of tight
tracking. The effort required to remain within the boundaries, especially with the more difficult models,
could easily produce occasional spikes of almost -2.0g, as seen in Figure 11. In the simulator instantaneous
motion was simulated using the motion system, but tight tracking was far less uncomfortable so the subjects
were not faced with trading comfort for performance.
14 of 19
(a) Aggressiveness vs duty cycle, illustrating different tracking (b) Aggressiveness increasing with boundary closure and task
“tactics”. difficulty.
(c) Duty cycle increasing with boundary closure and task dif- (d) Looking down the average slope.
ficulty.
Figure 14. Selected results from the BAT DART LAMARS data.
15 of 19
(c) Duty cycle increasing with boundary closure. (d) Looking down the average slope.
Figure 15. Selected results from the BAT DART NF-16D VISTA data.
particular models. This is clearly evidence of something already known; different flight control configurations
require different control “tactics.” The advantage of this method of characterizing inceptor workload comes
in describing the tactic. For instance, when comparing the types of tracking tactics employed in the VISTA
(see Figure 15(a)) it is clear that the Level Ib and III models required significantly higher duty cycles and
aggressiveness for successful tracking. The least successful subject was almost always using a tactic clearly
different than the other subjects in both the VISTA and LAMARS (specifically a much lower duty cycle)
showing how this method can clearly delineate between tracking strategies.
The subfigures (b) and (c) for both figures show provisional confirmation of the hypothesis that duty
cycle and aggressiveness will tend to increase as the pilot is required to work for better performance. By
using a first-order curve fit for each slope of aggressiveness and duty cycle to the logarithm of the boundary
displacement and assuming a gaussian distribution of slopes, the probability that aggressiveness and duty
cycle have an inverse relationship with boundary displacement can be estimated. The combination of the
two data sets give a probability of this correlation at approximately 90%. Subfigures 14(d) and 15(d) also
show this by being aligned with the view angle down the average slope.
16 of 19
17 of 19
XI. Conclusion
In the three years since the first conception of boundary avoidance tracking, the USAF Test Pilot School
has supported and conducted a series of flight and simulator research programs to further define and un-
derstand the phenomenon. These programs have largely supported the original boundary-avodince tracking
feedback model while showing that there is much to learn, especially with respect to how pilots transition
out of boundary avoidance tracking when the boundaries are exceeded. All experimental subjects have con-
firmed their conscious intent to avoid boundaries and the transition from point tracking to boundary tracking.
Boundary avoidance tracking has also been shown to be responsible for stable and unstable pilot-induced
oscillations.
In conjunction with this research, a new flight test technique that uses a series of collapsing boundaries
around a common task has proven a very useful tool for driving pilot gain, estimating the best attainable
closed-loop capability performance for a task, and identifying PIOs safely. The results from these tests
also appear to subjectively correlate well with Cooper-Harper tracking task results. The USAF TPS has
incorporated the collapsing boundary BAT FTT into its curriculum as an evolutionary change to the HQDT
technique. Early results are very promising; the intent of HQDT is met by giving pilots tasks that increase
their gain, not by asking that they maximize their gain.
It seems, then, that boundary avoidance tracking has circled back on point tracking and can be used
to characterize the pilot/aircraft capability for conducting a point tracking task. This should not be a
surprise. From the start of manned flights, pilots have recognized that no handling task can be accomplished
perfectly. “Good enough” has always instinctively been characterized as a certain maximum amount of error;
the boundary between success and failure.
Acknowledgments
The author’s continuing research into boundary avoidance tracking and its application in flight test would
not be possible without the generous support of the USAF Test Pilot School, including the enthusiastic
feedback and support of its talented staff and students.
18 of 19
Modeling for Flying Qualities Applications,” Tech. Rep. F33615-85-C-3610, Wright Research and Development Center Flight
Dynamics Laboratory, August 1989.
5 Twisdale, T. R. and Nelson, M. K., “A Method for the Flight Test Evaluation of PIO Susceptibility,” NASA Dryden
Tracking (HAVE BAT),” Tech. rep., United States Air Force Flight Test Center, Edwards AFB, CA, 2006.
11 Warren, R. D., An Investigation of the Effects of Boundary Avoidance on Pilot Tracking, Master’s thesis, US Air Force
Techniques, Master’s thesis, US Air Force Institute of Technology, Wright-Patterson AFB, OH, March 2007.
14 Cooper, G. E. and Harper, Jr., R. P., “The Use of Pilot Rating in the Evaluation of Aircraft Handling Qualities,” Tech.
19 of 19