Unmanned
Systems Autonomy and Automation
Autopilot
Time Dependencies
Shawn
Wyne
Embry-Riddle
Aeronautical University
October
4, 2015
Current
unmanned systems within the United States Air Force are typically operated
singly. That is, they are all treated as individual entities. Any cooperative
behavior is accomplished the same way manned aircraft do so. They share some
information, but the flight crews are directly responsible for any cooperation.
The increasing number of Remotely Piloted Aircraft (RPA) available mean that multiple
unmanned aircraft will be more frequently tasked to provide synergistic effects
on a common objective. One of the more difficult cooperative problems is that
of releasing weapons on specific targets. When multiple aircraft are involved,
it is possible to affect multiple weapons onto multiple distinct targets within
a finite area. However, in spite of multiple guidance types of weapons, there
is significant risk of weapons missing their target if the timing is not precise.
For manned aircraft, an acceptable tolerance for time-on-target is considered
within 30 seconds. But for the small weapon types and very specific target
types inherent to RPA operations, the margin of error is realistically plus or
minus two seconds. This is currently achievable, but only through significant
pilot effort. Because this coordination is difficult to achieve in practice, it
is not always attempted, even when it would be prudent. I propose the
development of a an autopilot module that will allow the RPA to, on its own,
affect a time restricted weapon impact at a specific location, while allowing a
full range of release parameters. The MQ-9 is a large RPA manufactured by
General Atomics-Aeronautical Systems, Inc. (GA-ASI). The current autopilot and
navigation system on the MQ-9 is not designed for these functions, and is not
intended for any weapon release at all, and certainly not one with time
restrictions. The module I propose must account for external environmental
parameters, user-defined weapon release parameters, weapon limitations,
aircraft performance capabilities and limitations, and most important of all, a
time constraint. The module will provide direct input to flight controls and
engine controls to maneuver the aircraft into the precise desired position at
the precise desired time, with a margin of error of less than two seconds.
Significance
United
States Air Force (USAF) doctrine identifies ten principles of war, some of
which are: offense, mass, economy of force, and surprise. Additionally, some of
the stated tenets of airpower are: concentration and synergistic effects
(United States Department of the Air Force AFDD-1, 2011). These principles and
tenets are the foundation of coordinated attack maneuvers that have been
practiced throughout modern military history. In the realm of air power, these
principles are met by placing kinetic weapon effects at a place to exert the
most harmful effect on the enemy. Communication and navigation has evolved to a
point that allows aircraft to be very precise in weapon placement. This
precision means that individual weapons can be placed onto very specific ground
locations. To achieve surprise and mass, multiple weapons can be placed in
unique locations at approximately the same time. The procedures a pilot, of an
RPA or manned aircraft, must follow can be very technical, and have little room
for error. Non-automated delivery systems in current RPA use rely on pilot
decision making for maneuver and execution. The decision process is best described
by John Boyd’s OODA loop (Tremblay, 2015). The cyclic process of “observe,
orient, decide, act” highlights the inherent difficulties in decision making
processes (Figure 1). But these processes, when understood, are simplified
through training and experience. Attempting to facilitate coordinated weapon
placement through RPA poses new challenges to the OODA loop. In particular,
even the “observe” step is complicated with excess data for a pilot to
assimilate. At the “decide” step, the level of precision is restricted to pilot
mental computational ability. A common training technique is to utilize
rules-of-thumb (ROT) to simplify decision making. ROT are inherently imprecise,
but for most purposes their accuracy is sufficient for the task. For example, a
timing ROT is to adjust indicated airspeed by one knot for every second of
timing error. On a one minute attack run, this correction will work to correct up
to around five seconds of error, but will be insufficient if the error is
greater. The entire process of the OODA loop and its limitations is
circumvented with strategic use of automation. Indeed, “The Air Force vision
for autonomy is to increase warfighter effectiveness by enhancing remotely
piloted systems capabilities and expanding their capacity to create effects in
the battlespace” (USAF RPA Vector, 2014, p. 40). In the case of timed weapon
attacks, the process is entirely mathematical. This makes it a perfect task for
a computer to handle with automation.
Alternative
With
some helpful tools, pilots are normally capable of placing weapons on target
within 30 seconds. However, RPA training only requires pilots to be proficient
to an accuracy of one minute (USAF 11-2MQ9v2, 2008). The maneuver problem for
timed weapon releases has several variables. Bombs are dropped from a ballistic
release point and gravity provides a fixed time of fall. The RPA must intercept
the release point at a precise time in order to meet the time on target.
Hellfire missiles, however, have a larger weapon engagement zone. But the
larger zone means the missile flight time is variable depending on where it is
released. The pilot must release the weapon in the zone, but only when the
current time of fall matches the needed impact time. Flying in a straight line
to the release point is the simplest maneuver. The MQ-9 ground control station
provides a timed measurement to a user created Control Point (GA-ASI, 2015).
Current groundspeed and distance to the point provide a countdown timer. The
pilot must make adjustments to speed in order to correct for deviations. But
this time is the time for the aircraft to reach the point, not a weapon on the
ground. In the case of a bomb, the ballistic fall path of the weapon is not
identical to the flight path of the aircraft, but it is close. Procedurally,
this can be used as a ROT to get close. The speeds involved only induce an
error of up to 5 seconds at the extreme (USAF AFTTP 3-3.MQ9, 2010). For more
precision, a chart is provided for pilots to adjust, mentally subtracting the
known error time from the planned time and using this as the new target. Part
of this chart is shown in Figure 2. Airspeed adjustments required to make
timing changes are also imprecise. At greater distances, small changes have
larger impacts. MQ-9 flies attack runs at 165 KTAS (USAF AFTTP 3-3.MQ9, 2010).
At 50 miles from a target, a five knot airspeed change yields a 35 second
change in target time. The process to adjust speed follows the OODA loop, but
each change requires a new loop to verify if the changes made are correct. Wind
also exacerbates the problem. The same speed change with a 60 knot tailwind
only changes the time by 18 seconds; a 60 knot headwind would make a change of 86
seconds. A long enough run provides enough time for a pilot to make continual
adjustments, evaluate, and update, until the time is exactly as needed. But the
tactical environment does not often allow for such an approach. The MQ-9
provides persistent full-motion video of a target area, normally at a range of
3 to 5 nautical miles (USAF AFTTP 3-3.MQ9, 2010). This new position means a
pilot must hold close to the target, and determine when to turn in for weapon
release. The control point method does not work here because the tool uses
instantaneous groundspeed for time calculations, so it is only accurate when
actually travelling directly at the point (GA-ASI, 2015). An aircraft turn
radius is determined by its speed and bank angle. At the standard 165 KIAS for
MQ-9 attacks, a standard rate turn has a radius of 0.9 NM, and turn time of 30
seconds. So a pilot must turn when the impact time minus fall time minus final
run in distance time minus turn time matches. If wind is present, the turn
distance can increase up to 1.5 NM or as little as 0.3 NM. Though seemingly
small, these changes in distance alter time to target by more than 1 minute
earlier or later (Figure 3). Since the hold position is slightly variable, and
the short time on final only allows for speed adjustments up to about five
seconds, this process is very difficult to do manually. But there is a tool to
automate some of the math. A program written in excel will perform the
calculations and present them to the pilot (Hosafros, 2012). The user inputs
some known variables and the program calculates the predicted turn times and
distances (Figure 4). This alleviates the mental math of various parameters,
but still requires the pilot to adjust position, cross-reference the tables,
decide when to turn, perform the turn at the correct moment, and make time
adjustments after the turn. Hellfire missiles also work with this program,
however pilot must choose a single release point instead of being able to use
the entire missile’s release envelope. The program is simple but does not
account for variable or gusting winds, time allocated to roll into and out of
turns, ground track errors on the pilot rollout, and it only estimates the
release point of a bomb. It also assumes the variables input by the pilot are
accurate, since it does not access data from the RPA itself. The tools
available make accurate timed weapon deliveries possible, but prone to human
error that makes success less certain.
Recommendation
The
MQ-9 has an existing on-board autopilot system. It uses an embedded GPS/INS and
typical pitot/static system to maintain stable commanded flight. For
redundancy, there are three of each autopilot component. The aircraft employs a
mid-level vote processor, which evaluates the data derived from each system to
determine which autopilot is best to use and to reject errors or failures
within the systems (GA-ASI, 2015). Manual
control for flight is typical, but the aircraft is capable of pre-programmed
flight. However, there are significant limitations to how the autopilot
processes pre-programmed commands. First, the system utilizes a fixed bank
angle of 14.5 degrees (GA-ASI, 2015). This is smaller than a standard rate
turn, and increases turn radius by as much as 50%. At a normal holding distance
of 5 NM, this shallow turn could place the aircraft with only 34 seconds
remaining until weapon release; not enough time to adjust for errors.
Pre-programmed flight also has no method to input desired times, and will not
change airspeed (GA-ASI, 2015). The autopilot needs a unique weapon specific
mode incorporated into the system to achieve the effects that a pilot can only
manually accomplish currently. The first adjustment needed is the ability to
manipulate bank angle up to 25 degrees as required to meet specific headings
and minimize turn radii. An increased roll rate will also facilitate accurate
maneuvers. The second adjustment is the ability to input a specific time on
target into the system. The autopilot keeps track of safe airspeed ranges
(GA-ASI, 2015). Since a normal attack is flown at 165 KTAS, this produces a
typical range of +/- 20 knots of adjustment allowed between Vne and Vstall. If
all the other parameters are monitored correctly, airspeed changes should not
be required. The next piece needed is a model to predict groundspeed at future
places in the aircraft flight path. From a hold perpendicular to the final run
axis, the current groundspeed does not match the future groundspeed after the
turn. Groundspeed is simply calculated as the addition of the measured wind
vector with the aircraft vector (Figure 5). The aircraft already contains a
wind sensor, which would be used for this purpose (GA-ASI, 2015). The next
element of the weapon autopilot is the turn estimator for time. The autopilot
already has a detailed flight model. It can be used to calculate and predict
the turn radius and time for any set of positional variables. The current manual
planning tool only outputs predicted turn times in hold distance increments of 0.5NM,
shown in the fourth column of Figure 4 (Hosafros, 2012). This works if the
aircraft happens to be at exactly one of those distances, but if not the pilot
must interpolate the correct time. The new autopilot will always be able to
precisely calculate from its exact position. Another element of the weapon
autopilot is control of the throttle to make fine adjustments to time. The
autopilot already controls throttle to maintain a commanded airspeed (GA-ASI,
2015). The new element will calculate the required airspeed to make time, and
use the new airspeed for its control logic. Engine performance responses to
throttle command are not instantaneous, so new airspeeds will not need to be
calculated at a very high rate. A calculation rate of 1/10 hz gives the engine
time to stabilize at the new commanded airspeed before attempting to adjust it
again. The final autopilot element that must be added is steering control for
weapon release. A bomb released from an aircraft follows a basic gravity
trajectory. The model is contained within the system already, and the target
location is backed up to the point of bomb release, called the CCRP
(Continuously Calculated Release Point) (General Atomics, 2015). It is
currently up to the pilot to fly to the CCRP, and also ensure the aircraft
ground track is directly pointed at the target. This ground track is the piece
the new autopilot must manage. Ground track, like ground speed, is simply
calculated. The autopilot will compare current ground track to desired ground
track, and adjust by commanding new headings to the autopilot.
A
more complicated aspect of this autopilot is its use for Hellfire missile
employment. The Hellfire is a powered weapon, which means it does not follow a
simple ballistic fall once released (Gleason Research Associates (GRA), Inc,
2015). Because the missile can also turn after release, it is not forced to be
used at a single CCRP. Figure 6 shows an example weapon engagement zone, where
a missile can be fired at any point within the green area (GRA, Inc, 2010).
Figure 6 shows the weapon fired 110 degrees to the right at a range of 9.8 km. Airspeed
and altitude both change the shape of the green engagement zone The time of
weapon fall in figure 6 is 53 seconds. Changing only the azimuth to 0 degrees
changes the weapon fall time to 38 seconds. This creates a more complicated
problem for the pilot to solve when attempting time constrained impacts. The
simplest solution is to eliminate the variable of azimuth. By turning the
problem into a single release point solution at an azimuth of 0, the same
planning tool can be utilized as for a bomb release. Even though this lets a
pilot meet time solutions, it negates the flexibility and usefulness of having
a weapon that can be fired off-azimuth. The newest Hellfire R variant has an
even greater turn ability to fire at targets almost completely behind the
aircraft (GRA, Inc, 2015). Current MQ-9 software contains no data for
calculating Hellfire weapon release parameters (GA-ASI, 2015). The weapon
flight characteristics must be included in the new autopilot module. A separate
program called P-Missile Impact Tool exists that will calculate release
parameters in real-time (GRA, Inc, 2010). This modeling should be used to not
only calculate current parameters, but also to predict future states. Within a
set of user-defined boundaries, the new autopilot module must use the same turn
prediction and airspeed regulation for time that the components use for bomb
dropping. It will add knowledge of acceptable release parameters to meet the
same end result.
The
new autopilot controls are not particularly difficult to implement. All the
math is two-dimensional linear geometry and trigonometry. The processors and
flight control surfaces are eminently capable of the speed and accuracy
required. But this type of autopilot system was never required before. Pilots
have been able to achieve the constraints presented manually. But the consequence
to this manual use is an artificial limitation on weapon envelopes and weapon
capabilities. Furthermore, the manual use is so difficult as to be infrequently
used even when time-restricted employment would be prudent. In order to make
the most significant impact in aerial warfare for the United States Air Force,
the MQ-9 autopilot needs to be updated for precise weapon engagement tactics.
General
Atomics Aeronautical Systems, Inc. (2015, August 4). Flight Manual, USAF Series
MQ-9 Aircraft, Serial Numbers 004, 006, 008, and Above. California: General Atomics – ASI.
Gleason
Research Associates, Inc. (2010). P Missile Impact Tool v1.1 [computer
software].
Gleason
Research Associates, Inc. (2015). Hellfire R: Fixed Wing. Retrieved from: https://www.grainc.net/capabilities/fixed.htm
Hosafros,
D. (2012). 17RS Planning Tool 6.0 [computer software].
Tremblay,
P. (2015, April 22). Shaping and Adapting: Unlocking the Power of Colonel John
Boyd’s OODA Loop. United States Marine
Corps Command and Staff College. Retrieved from: http://www.pogoarchives.org/straus/shaping-and-adapting-boyd-20150422.pdf
United
States Department of the Air Force. (2011, October 14). Air Force Doctrine
Document 1. Washington, DC: Headquarters, Department of the Air Force.
United
States Department of the Air Force. (2008, April 15). Air Force Instruction
11-2MQ-9 Volume 2: MQ-9 Crew Evaluation Criteria. Washington, DC: Headquarters,
Department of the Air Force.
United
States Department of the Air Force. (2010, September 15). Air Force Tactics,
Techniques, and Procedures 3-3.MQ9: Combat Aircraft Fundamentals MQ-9.
Washington, DC: Headquarters, Department of the Air Force.
United
States Department of the Air Force. (2014, February 17). RPA Vector: Vision and
Enabling Concepts 2013-2038. Washington, DC: Headquarters, Department of the
Air Force.
Appendix
Figure 1. OODA Loop diagram.
Reprinted from “Shaping and Adapting: Unlocking the Power of Colonel John
Boyd’s OODA Loop,” by P. Tremblay, 2015, United
States Marine Corps Command and Staff College.
Figure 2. Time Split Between
Control Point and Actual Time of Fall. From United States Department of the Air
Force. (2010, September 15). Air Force Tactics, Techniques, and Procedures
3-3.MQ9: Combat Aircraft Fundamentals MQ-9
Figure 3. Relative Turn Radii
and Time Differences for Changes in Wind
Figure
4.
17RS Planning Tool Output. D. Hosafros, 2012.
Figure
5.
Groundspeed Calculation Formula. From VirtualSkies, 2010, Aviation Navigation:
Calculations. NASA.
Figure
6.
Hellfire Weapon Engagement Zone. From Gleason Research Associates, Inc. (2010).
P Missile Impact Tool v1.1 [computer software].
