[American Institute of Aeronautics and Astronautics AIAA Infotech@Aerospace Conference - Seattle,...
Transcript of [American Institute of Aeronautics and Astronautics AIAA Infotech@Aerospace Conference - Seattle,...
American Institute of Aeronautics and Astronautics 092407
1
Fully Automated BMAV for Surveillance and
Reconnaissance on the Move
Ian D. Cowling, Simon. Willcox, Yoge. Patel and Phill R. Smith
Blue Bear Systems Research, Bedfordshire, UK. MK41 6JE.
The Blue Bear Systems Mini Air Vehicle (BMAV), is a fully automated unmanned air
vehicle. developed under the DTIC funded MAV program this vehicle was selected as the
low level vehicle in Team Stellar’s successful entry into the MOD’s Grand Challenge. Blue
Bear Systems Research have developed rapid prototyping technologies enabling bespoke
airframes to be developed in minimum time. The benefit of such an approach is the ability
to wrap an airframe around a payload depending on operator requirements. The direct
consequence of this is an ability to scale BMAV to meet user requirements. This paper
discusses the development of such a system, its bespoke subsystems and the development of
the fully automated algorithms.
I. Nomenclature
a = Free variable
x = X position
x0 = Initial x position
xf = Final x position
y = Y position
y0 = Initial y position
yf = Final y position
ψ = Heading angle
II. Introduction
HIS paper discusses the development of the fully automated Unmanned Air Vehicle BMAV. While UAVs offer
significant benefits to many different military and civilian organizations, further development is required to
reach the high standards in terms of efficiency offered by conventional air support [1]. The BMAV UAV system is
designed to significantly reduce operator workload through advanced control systems and algorithms and hence
improve efficiency.
One such demand on the operator is the set up time required per airframe before flight. The BMAV airframe is a
small portable UAV with a wing span of 1m. It can therefore be easily transported and either hand or catapult
launched. With an automatic launch mode, once launched the control law engages and progresses through a number
of control phases until the vehicle reaches a required height. After this point the vehicle can follow a predetermined
control sequence or can receive instructions from the ground control station. A catapult launch can be performed
with the catapult placed in the back of a small van, it is therefore possible to drive to the launch site and simply open
the doors before launching. This results in a negligible set up time.
The BMAV airframe itself is a completely scalable airframe. Rapid prototyping enables the wrapping of an airframe
around a payload to meet user requirements. By standardizing subsystems such as the BBSR Surveillance and
Navigation Autopilot (SNAP), the airframe can be sized according to requirements and therefore the airframe
specification is extremely flexible. For a standard 1m airframe with simple gimbaled camera the maximum flight
time is approximately 45 minutes. However, due to the scalable nature of the airframe this can be increased as well
as decreased to suit the demands of the operator.
After the completion of the mission BMAV switches to an advanced guidance algorithm which enables fully
automated landing in a constrained area. A good landing algorithm will bring the vehicle close to the desired target
T
AIAA Infotech@Aerospace Conference <br>and <br>AIAA Unmanned...Unlimited Conference 6 - 9 April 2009, Seattle, Washington
AIAA 2009-2072
Copyright © 2009 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
American Institute of Aeronautics and Astronautics 092407
2
in a reasonably gentle manner. As it descends, the control algorithm will attempt to reduce airspeed and direct the
vehicle to the target, however, the success of the algorithm is largely dependent on the approach which precedes this
stage [2]. If a soft landing is not a priority then potentially a higher degree of accuracy could be achieved by taking a
more aggressive landing strategy, however, typically a trade off is required to minimise damage to the airframe as
well as landing close to the target. In this case the landing accuracy is highly dependent on the initial guidance
algorithms which position the vehicle on the glide slope.
III. The BMAV airframe
The BMAV airframe is a flying wing constructed from lightweight EPP foam. There are a number of benefits of
this material including its lightweight construction, its resilience during transit as well as flight and its ease of repair
in the event of a heavy landing on hard ground. With a flight time of up to 45 minutes and with a potential payload
of up to 750g the BMAV is potentially capable of a range of surveillance and reconnaissance tasks. For surveillance
of a specific target the BMAV is fitted with a gimbaled camera, onboard video recording and a bespoke video link
with the ground control station.
Figure 1. The Blue Bear Mini Air Vehicle (BMAV).
IV. SNAP
The SNAP autopilot was initially developed in 2004 to aid the research and development carried out at Blue Bear
Systems Research (BBSR). BBSR specializes in rapid prototyping and therefore require an autopilot with a flexible
architecture and the ability to interface with bespoke and COTS hardware. The SNAP autopilot hosts control laws
developed in Simulink and compiled using the Real Time Workshop. The only requirement therefore for the control
law is a predefined list of signal inputs coming from the onboard sensors and additional hardware. The control laws
can therefore be designed and tested or tuned in flight. Since its development SNAP has been used in many BBSR
and external projects in UAVs ranging from 1kg through to 120kg.
Figure 2. BBSR hardware including a.) Amphibious UAV SNAP, b.) MIL SNAP, c.) Agentbox d.) MAV SNAP.
American Institute of Aeronautics and Astronautics 092407
3
V. Automated Take Off
Automatic take off from a catapult is triggered by SNAP sensing the acceleration along the x axis. The catapult can
provide up to 10g of acceleration momentarily which is more than can be typically encountered through
transportation and handling of BMAV, enabling a safe threshold for triggering the launch. The control algorithm
that is switched on after sensing the impulse is a simple wings level controller with the throttle and elevator
combining to achieve the maximum height rate climb. This is typically held for 5 seconds in which time the vehicle
will still be within 100m of the pilot and climbing steadily.
After this point to maintain visibility the vehicle is required to turn by switching to another flight mode, this can be
either waypoint tracking or a loiter pattern. Immediately after switching to the next phase it is essential that the
vehicle maintains height. This is ensured by limiting the bank angle demands to the controller, which in turn,
ensures that the vehicle maintains a sustainable turn rate at the expense of maneuverability. That is, the turning
circle will be much larger than if bank angle were not limited to preserve a degree of height hold.
A. Take off from inside a van
The catapult has been modified to ensure ease of use and minimal set up time. An electric winch mechanism has
been adopted which minimizes the set up time and an electric push button release enables the safety pilot to release
the vehicle while maintaining hands on the transmitter. The catapult itself can be set up outside or inside a van
providing cover for the operator and again reducing set up time. Figure 3 shows the BMAV in the top right hand
corner just after launch from inside the van. This figure shows the launch dolly separating from beneath the BMAV
as it leaves the rail.
Figure 3. Take off from inside a Transit van.
VI. Automated Recovery to a Constrained Area
An ideal landing algorithm will bring the vehicle close to the desired target and in a reasonably gentle manner. As it
descends, the control algorithm will attempt to reduce airspeed and direct the vehicle towards the target, however,
the success of the algorithm is largely dependent on the approach which precedes this stage. There are a number of
ways in which the vehicle can reduce height and airspeed. The simplest approach is to position the vehicle at the top
of a gradual glide slope and maintain a constant heading while reducing height and speed. Another approach could
be to loiter above the landing point while slowly reducing height and airspeed before transitioning into a landing
phase. If a soft landing is not a priority then potentially a higher degree of accuracy could be achieved by taking a
more aggressive landing approach, however, in this work it is desirable to minimize damage to the airframe. This
paper considers the glide slope approach to landing, as this is deemed more suitable for an urban environment where
a constrained landing may be required, for example, between buildings. The major challenge in this work is
therefore, the initial trajectory shaping in order to get the vehicle to the top of the glide slope with the correct
heading angle.
American Institute of Aeronautics and Astronautics 092407
4
Work carried out at the Naval Postgraduate School in California [2] presented a scheme for trajectory optimization
to solve the problem where a UAV is required to land on a ship. This work performs a non-linear trajectory
optimization to reach the top of the glide slope. This not only provides a feasible solution but also includes a
measure of optimality. Furthermore, the vehicles states (i.e. heading angle, position) are constrained at the initial and
terminal values to provide a smooth transition between control modes.
The drawback of this approach is the computational demand and therefore a simpler solution is demonstrated in this
work. By shaping a trajectory using a similar polynomial function, a feasible path can be found between any two
points, which guides the vehicle with the correct heading angle, however, in this case there is no measure of
optimality. The benefit of adopting this approach is that the initial and terminal states can be constrained analytically
to still provide this smooth transition between modes. This results in the initial heading angle of the reference
trajectory matching the heading angle of the vehicle at the time this control mode is switched on. Once the trajectory
is generated it is possible to split this path into a number of waypoints.
The guidance algorithm required to track closely spaced sequential waypoints is a challenging problem. However by
setting multiple waypoints the flight path is effectively more constrained and therefore if the vehicle is able to track
these waypoints it will do so within these constraints. This results in a greater control of not only the vehicle’s
destination but also the flight path shape. Controlling the flight path shape enables improved control of other states
such as roll angle and heading angle. Tracking of these closely spaced waypoints can be achieved through an
advanced integrated guidance and control solution such as preview control [3].
In order to land there are 3 phases considered in this work:
• Trajectory shaping; ensuring the vehicle reaches the top of the glide slope with the correct heading.
• Glide slope following, where significant height is dropped and the vehicle approaches the landing point along
a fixed heading.
• Landing; where the vehicle cuts throttle and maintains wings level flight until it lands.
The final phase is essentially open loop, as at this point, the guidance algorithms are switched off or limited, to
ensure the vehicle lands close to wings level. Therefore, it can be appreciated that in order to land in a constrained
area, this switch will be as late as possible and as close to the target as possible. It is therefore necessary to develop
an accurate guidance algorithm to prepare the vehicle for landing close to the target before switching into this final
mode. On the glide slope the vehicle drops height until it is close to the ground at which point there is a limited
amount of steering capability however, at this stage there is only a limited time until the vehicle must switch to the
final mode. Therefore, the first phase in which the vehicle approaches the glide slope is crucial in ensuring the
vehicle lands close to the desired target.
-300 -250 -200 -150 -100 -50 0 50 100 150-300
-250
-200
-150
-100
-50
0
50
100
150
East, (m)
Nort
h,
(m)
Target
100m circle
Flight path
Wind directionA
B C
Figure 4, Example landing flight path showing the different landing phases.
In order to land, the vehicle must first be down-wind of the landing target and heading towards it. Simply setting the
target as a waypoint is not sufficient as there is no control over which angle the waypoint is approached from.
American Institute of Aeronautics and Astronautics 092407
5
Therefore the first phase of the landing routine is to position the vehicle in the correct position with the correct
heading angle.
Figure 5. Landing preparation
By defining a landing axis, parallel to the wind direction, where 0x is perpendicular to the vehicles initial position.
Figure 6. The path shaping problem.
A polynomial can be used to shape the trajectory until the top of the glide slope.
20
xa
12
xa
6
xa
2
xa x a a
5
5
4
4
3
3
2
2
10 +++++=y
By constraining the initial and terminal positions of the trajectory and the derivatives the coefficients
(543210 ,,,,, aaaaaa ) can be found analytically. The derivatives are constrained to allow the initial and terminal
heading angles to be approximated:
=
0
0
0 tanx
ya
&
&ψ
=
f
f
fx
ya
&
&
tanψ
As the y axis is parallel to the wind axis the terminal heading angle is fixed to zero. By varying the initial heading
angle and position, a range of different trajectories are available as seen below.
American Institute of Aeronautics and Astronautics 092407
6
Figure 7. Path shaping for different initial positions and headings.
Figure 8. Path shaping for different initial positions and headings.
For a fully automated flight the vehicle must enter the first phase of the landing algorithm from a downwind
position. As the vehicle can be anywhere within range of the operator when the landing mode is engaged a sub-
phase is required before the landing algorithms are activated. This sub-phase consists of a simple semi-orbit function
which turns the vehicle towards to the landing point downwind of the target.
Figure 9 shows the landing trajectory of an automated BMAV flight carried out at the flying field near BBSR
premises. The flight trajectory is typical of the automated landing sequence whereby initially the vehicle flies wide
before aligning itself on the glide slope path. A steady reduction in height then brings the vehicle to its designated
landing area with wings level during the final stages.
Figure 9. Landing flight path.
American Institute of Aeronautics and Astronautics 092407
7
VII. Guidance cookbook for rapid automation
A. Waypoint following
The most common guidance mode is waypoint following. Although not necessarily providing the most stable of
guidance strategies for surveillance work, the waypoint following mode enables the vehicle to fly between points of
interest in minimal time. The guidance algorithm for this mode is trivial, calculating the required ground track based
on positional error and using GPS heading (ground track) to follow this flight path. Typically a constant altitude is
maintained during waypoint mode but each waypoint can have a different altitude demand if required.
The major problem with waypoint following is that although the waypoint itself is a fixed constraint, the flight path
in between the waypoints is not constrained. Instead, the flight path is very much dependent on the waypoint
positions relative to each other, the wind conditions and the closed loop dynamics of the vehicle.
For accurate flight path control a high density of waypoints or continuous path is required. The close spacing of
waypoints presents a potential problem of high gain oscillations within the inner loop controls. While these issues
are considered in specific manoeuvres such as the landing algorithm these are not covered in the waypoint following
mode. The problem is alleviated by appropriate tuning of the control gains.
Figure 10 show a typical waypoint following flight path over buildings near BBSR premises. Four waypoints are
specified in GPS co-ordinates. The BMAV attempts to reach each waypoint within a pre-specified radius around the
fixed co-ordinate. The flight paths are fairly consistent, and typically located within a tolerance band of 15m in
gusty conditions.
Figure 10. Waypoint following GPS trace and GCS interface.
B. High altitude loiter
A loiter pattern typically provides a more stable flight path for surveillance. Orbit, figure of eight and racetrack
loiter patterns are generally used for surveillance tasks. Orbits provide one of the most stable loiter patterns. In an
orbit loiter the vehicle essentially circles a specified point with a variable radius. The benefit of this loiter pattern is
not only in providing a stable platform but, once the UAV is orbiting, either a fixed or gimballed camera will focus
its footprint on the circle centre. In the absence of winds, this loiter pattern is fairly reliable. Thus, loiter patterns are
an excellent guidance strategy for surveillance of a static target. Another advantage of this guidance approach is that
the flight path shape is controlled, unlike the waypoint following mode, since the controller is demanding a
prescribed circular flight path. This regularity improves the predictability of the flight path and the repeatability in
the presence of winds.
American Institute of Aeronautics and Astronautics 092407
8
Figure 11. Loiter pattern example.
C. Dive attack and abort
The dive attack and abort function was initially developed to test and demonstrate UAV maneuverability and
controllability. From a high altitude loiter, the vehicle dives down at a given GPS location. As the vehicle
accelerates up to 30m/s vertically the wingtip camera is able to get a close up of the ground and the vehicle alerts
anybody in the area to the presence of the UAV. As the vehicle approaches the ground the vehicle pulls out of the
dive by demanding maximum elevator and throttle input. Figure 12 shows photo stills taken from the BMAV in a
dive, separated by 4 seconds of flight. The photos demonstrate the higher level of detail that is achievable, if
necessary, by vehicle maneuvering rather than by use of a more expensive and possibly heavier zoom lens camera.
Figure 12. Wing mounted onboard video capture during dive attack
D. Gimballed camera
Use of a gimballed camera is required to maintain visual lock on a target. The gimballed camera is fitted with two
independent servos controlling pan and tilt angles. It is connected to the SNAP autopilot which directs the camera at
the target. By considering the roll and pitch attitude of the vehicle the gimballed mechanism also removes large
oscillations from the image. To minimize damage on landing the camera can be retracted inside the fuselage on the
landing approach. Figure 13 shows typical camera footprint as the vehicle loiters above a static target (the gazebo in
this case) at a height of approximately 50m. The static target is kept well within view throughout the loiter pattern.
American Institute of Aeronautics and Astronautics 092407
9
Figure 13. Gimballed camera footage during surveillance flight.
VIII. The Ground Control Station
The BMAV vehicle can fly a preset mission or can receive instructions from the BBSR Ground Control Station
(GCS). The BBSR ground control station is built for complete flexibility and can take one of many different forms.
The only crucial component is the Snapserver which manages data transfer between the operator interfaces and
SNAP.
Figure 14. GCS communication links via Snapserver.
The operator interface can be any portable device which is capable of running JAVA. This device can be connected
directly to the telemetry unit and run the Snapserver as a background application or can be connected remotely to the
Snapserver. Snapserver itself handles the requests from the independent GCS components and provides the relevant
data to these applications. The GCS operator can therefore pick and choose the GCS components they want to view,
for example they may just want to view a moving map and specify waypoints or they may want to also view flight
data through plots or a traditional HUD. The benefit of this is the ability to meet the needs of a range of users from
researchers interested in specific state information through to end users who require a simplified command screen.
American Institute of Aeronautics and Astronautics 092407
10
Figure 15. Ground control station data flow.
Typically however the moving map will be present in the majority of ground control stations. This display shows the
latest vehicle position, recent vehicle trace, height information and any uploaded commands such as waypoints,
loiter points, landing positions and areas of interest. There is no limit to the number of points that can be uploaded,
however obviously the vehicle can only accept one flight mode at a time and therefore the GCS has the option to set
any particular mode to active. The landing position can be defined with or without the approach angle, this enables
the operator to specify the approach angle in the event of landing in a constrained area and therefore instructing the
vehicle to avoid any obstacles. If no approach angle is specified then the vehicle will approach the landing point into
wind which is calculated onboard using a wind estimation algorithm.
Figure 16. Path shaping for different initial positions and headings.
IX. Conclusion
This paper discusses the development of a scalable unmanned air system BMAV. Constructed from lightweight
EPP foam, BMAV is a completely scalable airframe which can be sized depending on user requirements and
payload specifications. Whilst maintaining a relatively simple airframe design, bespoke onboard systems and
advanced guidance algorithms minimize the operators workload. Fully automated flight can be initiated with the
press of a button and the vehicle can be recovered at a safe location due to the advanced automated landing
algorithm.
Future work will look to further develop the onboard systems and algorithms to include advanced features such
as area search patterns, moving target tracking, collision avoidance and multiple vehicle control.
American Institute of Aeronautics and Astronautics 092407
11
X. Acknowledgments
The authors would like to thank the Defence Technology and Innovation Centre (DTIC) for their support under
the project ‘Increasing Micro Air Vehicle Persistence in the Urban Environment ’
XI. References 1Johnson, C.W. “Act in haste, repent at leisure, an overview of operational incidents involving UAVs in Afghanistan,” 3rd
IET International conference on system safety, CP542, , 2008, 2Yakimenko, O. “Direct Method for Rapid Prototyping of Near Optimal Aircraft Trajectories” Journal of Guidance, Control
and Dynamics,vol 23. 2000, 3Farooq, A. and Limebeer, D.J.N. “Optimal trajectory tracking for missiles with Doppler beam sharpening radars” European
Control Conference, Kos, Greece. 2007,